Remote rehabilitation monitoring and intervention method and system based on smart wearable device

Abstract

The application provides a remote rehabilitation monitoring and intervention method and system based on a smart wearable device, comprising: extracting time domain features such as mean value and standard deviation and frequency domain features such as power spectral density from a clean physiological signal sequence, determining a feature vector for subsequent analysis; calculating a load change rate by combining historical motion data with a load state label, determining a critical point occurrence time if the change rate exceeds a preset threshold, and obtaining a critical point timestamp; updating remote rehabilitation model parameters according to symptom distress trigger factors, generating an intervention instruction sequence if the model parameters are adjusted and the predicted participation decreases, and obtaining an optimized intervention plan; transmitting the optimized intervention plan to the patient device in real time, adjusting the exercise intensity and duration, judging whether the patient feedback data is improved, and obtaining the final rehabilitation effect evaluation index.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, and in particular to a method and system for remote rehabilitation monitoring and intervention based on smart wearable devices. Background Technology

[0002] Pulmonary rehabilitation research for elderly patients with chronic obstructive pulmonary disease (COPD) is a crucial area of ​​health, directly impacting the quality of life and disease management outcomes in this population. With the increasing aging population, the incidence of this disease is rising annually, especially under remote rehabilitation models. Helping patients maintain exercise and alleviate symptoms has become a critical issue in healthcare services. Effective rehabilitation not only improves respiratory function but also reduces the risk of acute exacerbations, significantly contributing to the optimization of social medical resources.

[0003] However, current methods for monitoring and intervening in the exercise behavior of these patients still have significant shortcomings. Many programs tend to focus only on the patient's feedback after exercise, ignoring the real-time changes in the body's state and potential risks during exercise. This approach makes it difficult to capture the specific reasons why patients suddenly stop or avoid exercise, and it also fails to detect early signs of physical discomfort in a timely manner. As a result, the adjustment of rehabilitation plans lacks specificity, and patients may develop a fear of exercise as a result, affecting their long-term motivation to participate.

[0004] A deeper technical challenge lies in accurately identifying the critical point of physical load during exercise and the correlation between this load and symptoms such as shortness of breath. The critical point of physical load refers to the crucial moment when a patient's body may be unable to continue exercising due to shortness of breath or insufficient oxygen supply. If this critical state cannot be detected in time, it becomes difficult to determine whether the patient stops exercising due to physical limitations or psychological fear. Furthermore, identifying this critical point further impacts the understanding of symptom distress, as symptoms such as wheezing or fatigue are often the direct triggers for patients avoiding exercise. Without a precise grasp of the relationship between these two factors, rehabilitation guidance easily becomes superficial.

[0005] Therefore, how to capture the critical point of physical load during exercise in a remote environment in real time and link it to the mechanism of symptom distress has become a key issue in improving the rehabilitation effect of elderly patients with chronic obstructive pulmonary disease. Specifically, when patients are undergoing remotely guided exercise training at home, medical staff cannot intuitively determine the exact time when the patient stops exercising due to difficulty breathing, nor can they adjust the training intensity or provide psychological support based on these interruptions. A method is urgently needed to fill this gap. Summary of the Invention

[0006] This invention provides a remote rehabilitation monitoring and intervention method based on smart wearable devices, mainly including: Physiological signals of patients during remote rehabilitation are collected by wearable devices, including heart rate, respiratory rate and oxygen saturation data. Signal processing technology is used to filter out noise and obtain a clean physiological signal sequence. Based on the clean physiological signal sequence, time-domain features such as mean and standard deviation, and frequency-domain features such as power spectral density are extracted to determine the feature vector for subsequent analysis. The support vector machine algorithm is used to classify the feature vectors. If the classification result shows that the current physiological signal exceeds the preset normal range, it is judged as a potential increase in physical load, and a load status label is obtained. By combining load status tags with historical motion data, the load change rate is calculated. If the change rate exceeds a preset threshold, the time of the critical point is determined and the critical point timestamp is obtained. Obtain symptom log data corresponding to the critical point timestamps, including patients' self-reported wheezing and fatigue levels. Analyze the correlation strength between the logs and the workload using the random forest algorithm to determine the triggering factors of symptom distress. The parameters of the remote rehabilitation model are updated based on the triggering factors of symptom distress. If the predicted participation rate decreases after the model parameters are adjusted, an intervention instruction sequence is generated to obtain an optimized intervention plan. By optimizing the intervention plan and transmitting it to the patient's device in real time, adjusting the intensity and duration of exercise, and judging whether the patient's participation feedback data has improved, the final rehabilitation effect evaluation indicators are obtained.

[0007] This invention provides a remote rehabilitation monitoring and intervention system based on smart wearable devices, mainly comprising: The physiological signal acquisition and processing module is used to acquire physiological signals of patients during remote rehabilitation through wearable devices, including heart rate, respiratory rate and oxygen saturation data, and to use signal processing technology to filter out noise and obtain a clean physiological signal sequence. The feature extraction module is used to extract time-domain features, such as mean and standard deviation, and frequency-domain features, such as power spectral density, from the clean physiological signal sequence to determine the feature vector for subsequent analysis. The load status classification module is used to classify feature vectors using the support vector machine algorithm. If the classification result shows that the current physiological signal exceeds the preset normal range, it is judged as a potential increase in physical load and a load status label is obtained. The critical point determination module is used to calculate the load change rate by combining load status tags with historical motion data. If the change rate exceeds a preset threshold, the time when the critical point occurs is determined and the critical point timestamp is obtained. The symptom association analysis module is used to obtain symptom log data corresponding to the critical point timestamp, including the patient's self-reported wheezing and fatigue levels. The module uses a random forest algorithm to analyze the correlation strength between the logs and the workload to determine the triggering factors of symptom distress. The rehabilitation model update and intervention generation module is used to update the parameters of the remote rehabilitation model based on the triggering factors of symptom distress. If the predicted participation decreases after the model parameters are adjusted, an intervention instruction sequence is generated to obtain an optimized intervention plan. The intervention implementation and effect evaluation module is used to optimize the intervention plan and transmit it to the patient's device in real time, adjust the intensity and duration of exercise, determine whether the patient's participation feedback data has improved, and obtain the final rehabilitation effect evaluation indicators.

[0008] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a remote rehabilitation monitoring and intervention method based on smart wearable devices. Addressing the challenge of accurately determining the correlation between abnormal physiological signals, load changes, and symptom distress in remote rehabilitation, this method utilizes multi-dimensional data fusion and intelligent algorithms to resolve potential risks and personalized intervention needs during the rehabilitation process. The invention collects physiological signals such as heart rate, respiratory rate, and oxygen saturation in real time using wearable devices. Signal processing techniques are employed to filter noise, extract time and frequency domain features, and support vector machines are used to classify and determine load status. Combined with historical exercise data, the load change rate is calculated, and critical time stamps are located. Then, a random forest algorithm is used to analyze the correlation between symptom logs and load, identifying triggering factors. Finally, rehabilitation model parameters are dynamically updated, and an optimized intervention plan is generated, adjusting exercise intensity and duration in real time to ensure rehabilitation effectiveness. Through multi-algorithm collaboration and data-driven approaches, this invention achieves intelligent management of the entire process from signal monitoring to personalized intervention, significantly improving the safety and effectiveness of remote rehabilitation. Attached Figure Description

[0009] Figure 1 This is a flowchart of a remote rehabilitation monitoring and intervention method based on a smart wearable device according to the present invention.

[0010] Figure 2 This is a schematic diagram of a remote rehabilitation monitoring and intervention method based on a smart wearable device according to the present invention.

[0011] Figure 3 This is another schematic diagram of a remote rehabilitation monitoring and intervention method based on a smart wearable device according to the present invention.

[0012] Figure 4 This is a schematic diagram of the structure of a remote rehabilitation monitoring and intervention system based on a smart wearable device according to the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0014] like Figures 1-4This embodiment of a remote rehabilitation monitoring and intervention method and system based on smart wearable devices may specifically include: S101. Collect physiological signals of patients during remote rehabilitation through wearable devices, including heart rate, respiratory rate and oxygen saturation data, and use signal processing technology to filter out noise to obtain a clean physiological signal sequence.

[0015] Wearable devices were used to continuously monitor patients during remote rehabilitation, collecting physiological signal sequences including heart rate, respiratory rate, and oxygen saturation. Signal processing techniques were employed to preprocess the collected physiological signal sequences, using wavelet transform to decompose the signal components and filter out high-frequency noise and low-frequency baseline drift, resulting in a denoised signal sequence.

[0016] ; This represents the denoised values ​​of the wavelet detail coefficients. Represents the original wavelet detail coefficients. Indicates the threshold parameter. Represents a symbolic function. This represents the maximum value function, used to filter out high-frequency noise.

[0017] The denoised signal sequence is segmented according to time windows, and key physiological parameters are extracted from each segment to form a structured parameter set. If the parameter values ​​within certain time windows of the structured parameter set exceed a preset threshold range, the signal segment within that time window is subjected to secondary analysis to determine whether abnormal fluctuations exist and to obtain anomaly labeling results. Based on the anomaly labeling results, features are extracted from the signal segments marked as abnormal, and statistical indicators in the time and frequency domains are calculated to determine the specific type of abnormal signal. By combining the classification results of abnormal signal types with the patient monitoring background during remote rehabilitation, corresponding physiological state assessment data is generated to obtain the patient's real-time health status information. After obtaining the physiological state assessment data, a remote data transmission mechanism is automatically triggered based on the health status information reflected in the assessment data to store key information in a cloud database, completing the data archiving process.

[0018] When collecting physiological signals from patients during remote rehabilitation using wearable devices, smart bracelets or chest strap sensors can be used to record heart rate, respiratory rate, and blood oxygen saturation data in real time. For example, heart rate data can be collected 60 times per minute to obtain raw heart rate sequences such as 80, 82, 79, and 85 per minute. At the same time, respiratory rate can be collected 20 times per minute and blood oxygen saturation can be collected 10 times per minute. The data is transmitted to a cloud server via Bluetooth for storage and subsequent processing. Next, to address potential noise in the acquired raw signals, such as motion artifacts or environmental interference, signal processing techniques were employed for filtering. For example, wavelet transform was applied to the heart rate signal, with a decomposition level of 5 layers and the Daubechies wavelet basis function selected to extract low-frequency components and filter out high-frequency noise. This resulted in a smoothed heart rate sequence, such as 80, 81, 80, 82 bpm, reducing the noise amplitude from ±5 bpm to ±1 bpm. Simultaneously, a bandpass filter was used on the respiratory rate signal, with a passband range of 0.1 to 0.5 Hz, to remove low-frequency baseline drift and high-frequency interference. The filtered respiratory rate sequence changed from 18, 22, 19, 25 bpm to 18, 19, 19, 20 bpm, and the standard deviation decreased from 2.5 to 0.7. For blood oxygen saturation data, a moving average filter was used with a window size of 5 sampling points. The smoothed data changed from 95%, 94%, 96%, 93% to 95%, 95%, 95%, 94%, effectively reducing transient fluctuations. Finally, through the above processing, a clean physiological signal sequence is obtained, and this data is uploaded to the rehabilitation management system for correlation analysis with the patient's rehabilitation plan. For example, when the heart rate is stable at around 80 bpm, the system automatically determines that the patient is in a low-intensity rehabilitation phase. Combined with indicators such as a respiratory rate of 19 breaths / minute and blood oxygen saturation of 95%, the system generates exercise intensity suggestions that match the patient's condition and pushes them to the doctor for remote monitoring. This forms a complete technical chain from data acquisition to signal processing to rehabilitation assessment, ensuring data accuracy and rehabilitation effectiveness.

[0019] S102. Extract time-domain features, such as mean and standard deviation, and frequency-domain features, such as power spectral density, from the clean physiological signal sequence to determine the feature vector for subsequent analysis.

[0020] By performing hierarchical analysis on the clean sequence, the fluctuation patterns of physiological signals over different time periods are obtained, yielding preliminary signal distribution characteristics. Based on these characteristics, a segmented window technique is used to divide the signal, and time statistics are calculated for the data within each window to determine the mean and standard deviation of each segment. If the mean or standard deviation of a segment exceeds a preset threshold, deep frequency analysis is performed on that segment to obtain its power density distribution. By classifying the power density distribution, frequency components related to abnormal fluctuations are extracted to identify potential abnormal signal segments. For these abnormal signal segments, a support vector machine algorithm is used to classify the feature vectors and determine whether the signal fluctuations belong to a specific physiological state.

[0021] ; Indicates the classification result. Represents a symbolic function. Indicates the number of support vectors. Represents the Lagrange multipliers. Represents support vector labels. Represents the kernel function. Represents support vector features, This represents the input feature vector. This represents the bias, used in support vector machines to classify and determine physiological states.

[0022] Based on the classification results, comprehensive indicators related to physiological state are obtained, generating structured state data records. These structured state data records are then transmitted to the cloud for remote storage and backup, resulting in a persistent signal analysis archive.

[0023] In the field of remote rehabilitation, when extracting features from cleaned physiological signal sequences, a comprehensive analysis process can be achieved through information technology. Firstly, for the extraction of time-domain features, the system automatically calculates the average and standard deviation of the heart rate signal within a 10-minute window. For example, processing the cleaned heart rate sequence data of 75, 78, 76, 77, and 79 bpm yields an average of 77 bpm and a standard deviation of 1.3, reflecting the signal's central tendency and volatility. Simultaneously, analyzing respiratory rate sequences such as 17, 18, 17, 19, and 18 breaths per minute yields an average of 17.8 breaths per minute and a standard deviation of 0.8, demonstrating respiratory stability. Subsequently, for frequency domain feature extraction, the system uses the Fast Fourier Transform algorithm to perform power spectral density analysis on the heart rate signal. With a sampling frequency of 1 Hz, the calculated dominant frequency component is at 0.2 Hz, with a power spectral density value of 3.5, indicating that the main periodic changes in the heart rate signal are concentrated in this frequency range. Frequency domain analysis of the respiratory rate signal shows a dominant frequency of 0.3 Hz and a power spectral density value of 2.8, reflecting the regularity of the respiratory cycle. Finally, the system integrates the above time-domain and frequency-domain features into feature vectors. For example, the heart rate feature vector is [77, 1.3, 0.2, 3.5], and the respiratory rate feature vector is [17.8, 0.8, 0.3, 2.8]. These vectors are stored in a cloud database, linked to the patient's rehabilitation records, and automatically generate feature analysis reports for subsequent rehabilitation status assessment models. This forms a complete logical chain from signal feature extraction to data integration, ensuring the systematic nature of the analysis process and the traceability of the data.

[0024] S103. The support vector machine algorithm is used to classify the feature vectors. If the classification result shows that the current physiological signal exceeds the preset normal range, it is judged as a potential increase in physical load, and a load status label is obtained.

[0025] The collected physiological signals undergo preliminary cleaning to obtain interference-free signal data and determine the basic signal sequence. Based on the basic signal sequence, signal segments within different time windows are extracted using a time-segmented extraction method, resulting in segmented signal sets. For each segmented signal set, the signal fluctuation amplitude within each time window is calculated. If the fluctuation amplitude exceeds a preset threshold range, it is marked as a high-fluctuation segment, resulting in a marked signal segment. The feature vectors in the marked signal segments are classified using a support vector machine algorithm. If the classification result shows that the signal exceeds a preset normal range, it is determined to be a potential load increase, resulting in a load status label. Based on the load status label, the temporal distribution information related to the high-fluctuation segment is obtained to determine the specific time period of load increase. By performing a deep comparison of the signal data during the load increase time period, the differences between the signals and signals from other time periods are obtained to determine the persistence characteristics of the load increase.

[0026] ; This indicates the difference in average signal strength between periods of increased load and other periods. Indicates the number of comparison time points. This represents the signal value at time t during the load increase period. This represents the signal value at time t during the normal segment, used to determine the persistence characteristic.

[0027] To address the persistent nature of the load, data storage technology is used to archive relevant information, resulting in structured load status records.

[0028] In the field of remote rehabilitation, the system utilizes information technology to classify and assess the load status of physiological signal feature vectors, constructing a complete analysis process. First, the system inputs the extracted feature vectors into a support vector machine (SVM) algorithm for classification. For example, for a patient's post-exercise heart rate feature vector [82, 2.5, 0.25, 4.2], the kernel function of the SVM model is set to the radial basis function (RBF), automatically calculating the classification boundary. The result shows that the distance between this vector and the preset normal range is 1.8, exceeding the preset threshold of 1.0, indicating a preliminary judgment of an abnormal state. Simultaneously, a similar analysis is performed on the respiratory rate feature vector [19.5, 1.2, 0.35, 3.1], and the calculation results show a distance of 1.5, also exceeding the threshold, indicating a possible increase in physical load. Subsequently, the system performs comparative analysis with historical data, calling upon the patient's average heart rate feature vector [78, 1.8, 0.22, 3.8] and respiratory rate feature vector [17.2, 0.9, 0.32, 2.9] from the past week. Using the Euclidean distance algorithm, the deviations between the current data and historical data are calculated to be 2.3 and 2.1 respectively, confirming that the current signal deviates from the normal trend and further verifying the possibility of increased workload. Finally, based on the classification results and deviation analysis, the system automatically generates a workload status label, for example, marking the current status as "moderate workload increase," and associates this label with the feature vector data, storing it in the cloud-based rehabilitation record for subsequent rehabilitation plan adjustment modules to access. This forms a complete logical chain from classification to status assessment, ensuring data continuity and analytical reliability.

[0029] S104. Calculate the load change rate by combining the load status label with historical motion data. If the change rate exceeds the preset threshold, determine the time when the critical point occurs and obtain the critical point timestamp.

[0030] By using load status tags, corresponding time period information is extracted from historical motion data. Data matching technology is employed to obtain the motion intensity distribution related to load status, determining a preliminary load change trend. For this trend, the rate of change within different time periods is calculated. A time window approach is used to obtain rate data within each window, resulting in a rate of change sequence. If the rate within a time window exceeds a preset threshold, it is marked as an abnormal rate segment. Combined with historical motion data, the existence of critical points is determined, resulting in an abnormal segment identifier. For each abnormal segment identifier, the corresponding occurrence time information is obtained. Timestamp technology is used to associate the abnormal rate segment with a specific time, determining the precise timestamp of the critical point. Based on the timestamp of the critical point, load status data for the time periods before and after it is extracted. Data comparison technology is used to analyze the fluctuations in load status before and after the critical point, obtaining a status fluctuation record. For this status fluctuation record, data archiving technology is used to structure and store the critical point, occurrence time, and relevant load change information, obtaining a complete load change archive.

[0031] In the field of remote rehabilitation, the system uses information technology to conduct in-depth analysis of patients' workload status tags and historical movement data to identify key moments of change in their physical condition.

[0032] For example, the system first extracts a patient's workload status label data for the past month from the cloud-based rehabilitation records. Assuming the current label is "high workload status", the corresponding workload index is 85.5, while the average workload index for the past 30 days is 70.2. The system calculates the workload change rate using a linear regression algorithm, and the change rate is (85.5-70.2) / 30=0.51 / day. Then, the change rate is compared with the preset threshold of 0.3 / day, and it is found that it has significantly exceeded the threshold, indicating that the workload status has abnormal fluctuations. Next, the system further retrieves historical exercise data, extracting the patient's exercise duration and intensity records for the past 7 days. For example, the most recent day's exercise duration was 45 minutes, with an intensity score of 7.2, while the average exercise duration for the previous 6 days was 30 minutes, with an intensity score of 5.8. Using a weighted average algorithm, the exercise load increment is calculated as (45-30)*0.4 + (7.2-5.8)*0.6 = 6.84. Combined with the load change rate, the system pinpoints the turning point when the load index rises from 70.2 to 85.5, accurate to the specific date and hour, assumed to be 14:30 on October 15, 2023, generating a critical point timestamp of "202310151430". Simultaneously, the system associates this timestamp with the exercise data and automatically stores it in the patient's file, forming a complete analytical chain from load change rate calculation to critical point determination. This ensures that the subsequent rehabilitation plan adjustment module can accurately access this data to optimize intervention timing.

[0033] S105. Obtain symptom log data corresponding to the critical point timestamp, including the patient's self-reported wheezing and fatigue levels. Analyze the correlation strength between the logs and the workload using the random forest algorithm to determine the triggering factors of symptom distress.

[0034] Step 1: Starting from the critical timestamp, obtain the corresponding symptom log records, including patient-reported wheezing and fatigue levels, and organize them into a structured dataset to obtain a preliminary set of symptom information. Step 2: From the preliminary symptom information set, extract records related to workload data. Using data matching techniques, align symptom manifestations with workload fluctuations in time to determine the time-related dataset. Step 3: Based on the time-related dataset, use the random forest algorithm to calculate the correlation strength between symptom manifestations and workload fluctuations, analyze the interaction patterns between the two near the critical timestamp, and obtain the correlation strength results.

[0035] ; Indicates the strength of the association between symptoms and workload. This represents the number of trees in a random forest. Indicates the first The reduction in Gini impurity in trees with symptom and load characteristics.

[0036] Step 4: Based on the correlation strength results, if the correlation strength between a certain symptom and the load data exceeds a preset threshold, it is marked as a potential triggering factor. The corresponding symptom log fragment is then retrieved to identify the key triggering factor. Step 5: Based on the key triggering factor, load data and symptom records for the relevant time period are extracted. Data comparison techniques are used to analyze the changing trends of both within different time windows to determine the symptom triggering pattern. Step 6: For the symptom triggering pattern, relevant information such as critical timestamps, load fluctuations, and symptom manifestations is archived and stored in a database using structured storage technology to obtain a complete symptom load profile.

[0037] In the field of remote rehabilitation, the system uses information technology to conduct in-depth analysis of patients' symptom log data to explore the potential link between load changes and physical discomfort. First, the system extracts patient symptom log data from the cloud database corresponding to the critical time stamp "202310151430". Assuming that within 24 hours before and after this time point, the patient's self-reported wheezing score was 6.8 (out of 10) and fatigue score was 7.5 (out of 10), while the average wheezing score over the past 7 days was 4.2 and the fatigue score was 5.0, the system automatically calculates the symptom increment, resulting in a wheezing increment of 6.8 - 4.2 = 2.6 and a fatigue increment of 7.5 - 5.0 = 2.5. Subsequently, the system invoked the random forest algorithm to perform correlation analysis between symptom log data and concurrent load index data (assuming a critical load index of 82.3 and a 7-day average load index of 68.5). Through feature importance assessment, the contribution of the load index to the wheezing score was calculated to be 0.65, and its contribution to the fatigue score to be 0.72, indicating a strong correlation between load changes and symptom exacerbation. Next, the system further extracted environmental variable data as an auxiliary analysis dimension. For example, the temperature on the critical day was 18 degrees Celsius and the humidity was 75%, while the average temperature over the past 7 days was 22 degrees Celsius and the humidity was 60%. Algorithm analysis showed that the indirect impact coefficient of humidity changes on the wheezing score was 0.15. Considering both load and environmental factors, the system ultimately determined that increased load was the main trigger for symptom distress, with a correlation strength score of 0.68 (threshold 0.5). Finally, the system automatically archived the analysis results into the patient's rehabilitation record, forming a complete data chain from symptom log extraction to trigger factor identification, providing a basis for subsequent rehabilitation strategy optimization.

[0038] S106. Update the parameters of the remote rehabilitation model based on the triggering factors of symptom distress. If the predicted participation decreases after the model parameters are adjusted, generate an intervention instruction sequence to obtain an optimized intervention plan.

[0039] Step 1: Obtain trigger factor data related to symptom distress from the database, classify and process this data, and organize the frequency of occurrence of trigger factors and corresponding symptom manifestations in chronological order to obtain a set of classified trigger factors.

[0040] ; This indicates the frequency of occurrence of the k-th type of triggering factor. This indicates the total number of observation time points. Indicates the first Triggering factors at each point in time, Indicates an indicator function, when equal The value is 1 if the time condition is not met and 0 otherwise. This formula is used to classify and process trigger factor data to obtain a frequency set.

[0041] Step Two: Based on the categorized set of triggering factors, extract historical records related to remote rehabilitation. Use data matching technology to associate the triggering factors with the operating parameters of the rehabilitation model, determining the correspondence between parameters and symptom distress. Step Three: Analyze and predict the trend of participation changes based on the correspondence between parameters and symptom distress. If the predicted participation value is lower than a preset threshold, mark it as a declining participation state and obtain the corresponding parameter adjustment requirements. Step Four: Based on the parameter adjustment requirements in the declining participation state, generate intervention instructions. Use serialization technology to arrange the intervention instructions by priority, determining an ordered instruction sequence. Step Five: For the ordered instruction sequence, combined with the current state of the remote rehabilitation model, adjust the model parameters, generate the specific content of the optimized intervention, and obtain the adjusted intervention plan. Step Six: Based on the adjusted intervention plan, associate and store the intervention instruction sequence with the triggering factor data, using structured storage technology to save it to the database, determining the final intervention execution basis. Step Seven: Based on the final intervention execution basis, periodically extract the execution records of the intervention plan from the database, analyze the changes in symptom distress through data comparison technology, and obtain an updated symptom management file.

[0042] In the field of remote rehabilitation, the system analyzes the triggering factors of patients' symptoms using information technology and automatically updates rehabilitation model parameters to optimize intervention strategies. First, based on identified symptom triggering factors, such as abnormal heart rate caused by excessive exercise load, the system extracts the patient's heart rate data from the past 14 days. Assuming the current critical heart rate is 98 beats per minute and the average heart rate over the past 14 days is 82 beats per minute, the system calculates the heart rate increment as 98 - 82 = 16 beats per minute. Next, the system uses a support vector machine algorithm to perform regression analysis on the heart rate increment and exercise load data. Assuming the critical exercise load is 75.5 units and the 14-day average load is 62.0 units, the analysis shows that the weight of exercise load on the heart rate increment is 0.78, indicating that load is the main influencing factor. Subsequently, the system automatically adjusts the rehabilitation model parameters based on the analysis results, lowering the upper limit of exercise load from 75.5 units to 70.0 units and re-predicting the patient's participation rate for the next 7 days. Assuming the predicted participation rate was 85% before the adjustment and decreased to 78% after the adjustment, it falls below the preset threshold of 80%. To this end, the system automatically triggers an intervention instruction generation mechanism, optimizes the intervention sequence through a genetic algorithm, and calculates the optimal intervention plan. This includes reducing daily exercise time from 60 minutes to 45 minutes, increasing weekly rest days from 2 to 3, and increasing heart rate monitoring frequency from once daily to twice daily. Finally, the system synchronizes the optimized intervention plan and adjustment parameters to the cloud-based rehabilitation record, forming a complete data loop from trigger factor analysis to intervention optimization, providing support for subsequent personalized adjustments.

[0043] S107. By optimizing the intervention plan and transmitting it to the patient's device in real time, adjusting the intensity and duration of exercise, judging whether the patient's participation feedback data has improved, and obtaining the final rehabilitation effect evaluation index.

[0044] Step 1: Retrieve the latest optimized plan data from the intervention plan database. Using automated transmission technology, send the plan content to the patient's device via a real-time transmission channel, ensuring the integrity of the transmission process. Step 2: For the intervention plan transmitted to the patient's device, analyze the exercise intensity and duration parameters. Dynamically adjust the exercise task using the device's built-in control module to obtain the adjusted task execution instructions. Step 3: Based on the adjusted task execution instructions, record the feedback data from the patient's device during execution. Use data acquisition technology to store the feedback data in real-time, ensuring a complete record. Step 4: For the complete record of feedback data, use data analysis tools to classify it. If key indicators in the feedback data are below a preset threshold, mark it as needing optimization, obtaining the classified data analysis results. Step 5: Based on the classified data analysis results, extract the feedback data in the need-to-optimize state. Compare it with the effect evaluation criteria in the intervention plan to determine if there is any deviation and identify the specific deviation range. Step 6: For the specific deviation range, generate corresponding adjustment plan data. Use automated update technology to transmit the adjustment plan to the patient's device, obtaining the updated exercise task configuration. Step 7: Based on the updated exercise task configuration, continuously monitor changes in the patient's device feedback data, track the trends using data analysis tools, and determine the final assessment criteria for rehabilitation effectiveness.

[0045] In the field of remote rehabilitation, the system uses information technology to achieve real-time transmission and dynamic adjustment of intervention plans, and analyzes patient feedback data to evaluate rehabilitation effects. First, the system transmits the optimized intervention plan to the patient's smart wearable device via a cloud server. Assuming the plan adjusts the daily exercise intensity from 80% to 65% and the exercise duration from 50 minutes to 35 minutes, encryption protocols are used during transmission to ensure data security. Next, the device automatically updates the patient's exercise tasks according to the plan and collects feedback data in real time through built-in sensors. For example, if the patient's average cadence over the past 7 days was 110 steps per minute, and the cadence stabilized at 95 steps per minute for 3 days after the adjustment, the system calculates a cadence reduction of 110-95=15 steps / minute, indicating that the exercise intensity adjustment has taken effect. Subsequently, the system uses a decision tree algorithm to classify and analyze the feedback data, assessing the patient's physical tolerance. Assuming the tolerance score increases from 72 points before the adjustment to 81 points after the adjustment, exceeding the preset threshold of 75 points, and combined with the data showing an increase in daily exercise completion rate from 85% to 92%, the system concludes that the patient's participation feedback has significantly improved. To further refine the assessment, the system introduces rehabilitation effectiveness evaluation indicators, combining tolerance scores and completion rates. A weighted average algorithm is used to calculate the final effectiveness index. Assuming a tolerance weight of 0.6 and a completion rate weight of 0.4, the effectiveness index is calculated as 81 × 0.6 + 92 × 0.4 = 85.4 points, higher than the target value of 80 points, indicating good rehabilitation effectiveness. Finally, the system automatically archives the evaluation results to a cloud database and dynamically adjusts the parameters of subsequent intervention plans based on the effectiveness index, forming a complete closed loop from plan transmission to effectiveness evaluation, providing continuous support for personalized rehabilitation.

[0046] This invention provides a remote rehabilitation monitoring and intervention system based on smart wearable devices, mainly comprising: The physiological signal acquisition and processing module is used to acquire physiological signals of patients during remote rehabilitation through wearable devices, including heart rate, respiratory rate and oxygen saturation data, and to use signal processing technology to filter out noise and obtain a clean physiological signal sequence. The feature extraction module is used to extract time-domain features, such as mean and standard deviation, and frequency-domain features, such as power spectral density, from the clean physiological signal sequence to determine the feature vector for subsequent analysis. The load status classification module is used to classify feature vectors using the support vector machine algorithm. If the classification result shows that the current physiological signal exceeds the preset normal range, it is judged as a potential increase in physical load and a load status label is obtained. The critical point determination module is used to calculate the load change rate by combining load status tags with historical motion data. If the change rate exceeds a preset threshold, the time when the critical point occurs is determined and the critical point timestamp is obtained. The symptom association analysis module is used to obtain symptom log data corresponding to the critical point timestamp, including the patient's self-reported wheezing and fatigue levels. The module uses a random forest algorithm to analyze the correlation strength between the logs and the workload to determine the triggering factors of symptom distress. The rehabilitation model update and intervention generation module is used to update the parameters of the remote rehabilitation model based on the triggering factors of symptom distress. If the predicted participation decreases after the model parameters are adjusted, an intervention instruction sequence is generated to obtain an optimized intervention plan. The intervention implementation and effect evaluation module is used to optimize the intervention plan and transmit it to the patient's device in real time, adjust the intensity and duration of exercise, determine whether the patient's participation feedback data has improved, and obtain the final rehabilitation effect evaluation indicators.

[0047] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A remote rehabilitation monitoring and intervention method based on smart wearable devices, characterized in that, The method includes: Physiological signals of patients during remote rehabilitation are collected by wearable devices, including heart rate, respiratory rate and oxygen saturation data. Signal processing technology is used to filter out noise and obtain a clean physiological signal sequence. Based on the clean physiological signal sequence, time-domain features such as mean and standard deviation, and frequency-domain features such as power spectral density are extracted to determine the feature vector for subsequent analysis. The support vector machine algorithm is used to classify the feature vectors. If the classification result shows that the current physiological signal exceeds the preset normal range, it is judged as a potential increase in physical load, and a load status label is obtained. By combining load status tags with historical motion data, the load change rate is calculated. If the change rate exceeds a preset threshold, the time of the critical point is determined and the critical point timestamp is obtained. Obtain symptom log data corresponding to the critical point timestamps, including patients' self-reported wheezing and fatigue levels. Analyze the correlation strength between the logs and the workload using the random forest algorithm to determine the triggering factors of symptom distress. The parameters of the remote rehabilitation model are updated based on the triggering factors of symptom distress. If the predicted participation rate decreases after the model parameters are adjusted, an intervention instruction sequence is generated to obtain an optimized intervention plan. By optimizing the intervention plan and transmitting it to the patient's device in real time, adjusting the intensity and duration of exercise, and judging whether the patient's participation feedback data has improved, the final rehabilitation effect evaluation indicators are obtained.

2. The remote rehabilitation monitoring and intervention method based on a smart wearable device according to claim 1, characterized in that, The process involves collecting physiological signals from the patient during remote rehabilitation via wearable devices, including heart rate, respiratory rate, and oxygen saturation data. Signal processing techniques are then used to filter out noise, resulting in a clean physiological signal sequence, including: Continuous monitoring of patients during remote rehabilitation using wearable devices, collecting physiological signal sequences including heart rate data, respiratory rate, and oxygen saturation; Signal processing techniques were used to preprocess the acquired physiological signal sequences. Wavelet transform was applied to decompose the signal components, filter out high-frequency noise and low-frequency baseline drift, and obtain the denoised signal sequence. ； This represents the denoised values ​​of the wavelet detail coefficients. Represents the original wavelet detail coefficients. Indicates the threshold parameter. Represents a symbolic function. This represents the maximum value function, used to filter out high-frequency noise; The denoised signal sequence is segmented, divided into segments according to time windows, and key physiological parameters are extracted from each segment to form a structured parameter set. If the parameter values ​​in certain time windows of the structured parameter set exceed the preset threshold range, then the signal segment within that time window is subjected to secondary analysis to determine whether there are abnormal fluctuations and to obtain the abnormality marking results. Based on the anomaly labeling results, feature extraction is performed on the signal segments labeled as anomalies, statistical indices in the time and frequency domains are calculated, and the specific type of the anomaly signal is determined. By classifying the abnormal signal types and combining the patient monitoring background during remote rehabilitation, corresponding physiological status assessment data is generated to obtain the patient's real-time health status information. After acquiring physiological status assessment data, the remote data transmission mechanism is automatically triggered based on the health status information reflected in the assessment data, storing key information in the cloud database and completing the data archiving process.

3. The remote rehabilitation monitoring and intervention method based on intelligent wearable devices according to claim 1, characterized in that, The step of extracting time-domain features, such as mean and standard deviation, and frequency-domain features, such as power spectral density, from clean physiological signal sequences to determine feature vectors for subsequent analysis includes: By performing hierarchical analysis on the clean sequence, the fluctuation patterns of physiological signals in different time periods were obtained, and preliminary signal distribution characteristics were obtained. Based on the preliminary signal distribution characteristics, the signal is segmented using a segmented window technique. Time statistics are calculated for the data within each window to determine the average value and standard deviation of each segment. If the average value or standard deviation of a certain segment exceeds the preset threshold range, then a deep frequency analysis is performed on the signal of that segment to obtain its power density distribution. By classifying the power density distribution, frequency components related to abnormal fluctuations are extracted to identify potential abnormal signal segments. For abnormal signal segments, the support vector machine algorithm is used to classify the feature vectors to determine whether the signal fluctuations belong to a specific physiological state. ； Indicates the classification result. Represents a symbolic function. Indicates the number of support vectors. Represents the Lagrange multipliers. Represents support vector labels. Represents the kernel function. Represents support vector features, This represents the input feature vector. This represents the bias, used in support vector machines to classify and determine physiological states. Based on the classification results, comprehensive indicators related to physiological state are obtained, and structured state data records are generated. By transmitting structured state data records to the cloud, remote storage and backup of the data are achieved, resulting in a persistent signal analysis archive.

4. The remote rehabilitation monitoring and intervention method based on a smart wearable device according to claim 1, characterized in that, The feature vectors are classified using a support vector machine algorithm. If the classification result shows that the current physiological signal exceeds a preset normal range, it is judged as a potential increase in physical burden, and a burden status label is obtained, including: By performing preliminary cleaning and processing on the collected physiological signals, the signal data after removing interference is obtained, and the basic signal sequence is determined. Based on the basic signal sequence, signal segments within different time windows are obtained by using a time-segmented extraction method to obtain a segmented signal set. For a segmented signal set, the signal fluctuation amplitude within each time window is calculated. If the fluctuation amplitude exceeds a preset threshold range, it is marked as a high fluctuation segment, and the marked signal segments are obtained. The support vector machine algorithm is used to classify the feature vectors in the labeled signal segments. If the classification result shows that the signal exceeds the preset normal range, it is determined that the potential load has increased, and the load status label is obtained. Based on the load status labels, obtain the time distribution information related to high fluctuation segments to determine the specific time periods of load increase; By performing in-depth comparison of signal data during periods of increased load, we can obtain information on the differences between signals from other time periods and determine the persistence characteristics of the increased load. ； This indicates the difference in average signal strength between periods of increased load and other periods. Indicates the number of comparison time points. This represents the signal value at time t during the load increase period. This represents the signal value at time t during the normal segment, used to determine persistence characteristics; To address the persistent nature of the load, data storage technology is used to archive relevant information, resulting in structured load status records.

5. The remote rehabilitation monitoring and intervention method based on a smart wearable device according to claim 1, characterized in that, The process involves calculating the load change rate by combining load status tags with historical motion data. If the change rate exceeds a preset threshold, the time of the critical point is determined, and a critical point timestamp is obtained. This includes: By using load status labels, corresponding time period information is extracted from historical motion data. Data matching technology is then used to obtain the motion intensity distribution related to load status and determine the preliminary load change trend. To analyze the load change trend, the rate of change within different time periods is calculated. By dividing the time window, the rate data within each window is obtained to generate a rate of change sequence. Based on the rate of change sequence, if the rate exceeds a preset threshold within a certain time window, it is marked as an abnormal rate segment. Combined with historical motion data, it is determined whether there is a critical point and an abnormal segment identifier is obtained. For the abnormal segment identifier, the corresponding occurrence time information is obtained. Through timestamp technology, the abnormal rate segment is associated with the specific time to determine the precise timestamp of the critical point. Based on the timestamp of the critical point, load status data for the time periods before and after it are extracted. Data comparison technology is used to analyze the fluctuation of load status before and after the critical point, and a status fluctuation record is obtained. For status fluctuation records, data archiving technology is used to structure and store information related to critical points, their occurrence times, and load changes to obtain a complete load change archive.

6. The remote rehabilitation monitoring and intervention method based on a smart wearable device according to claim 1, characterized in that, The process involves acquiring symptom log data corresponding to the critical point timestamps, including patient-reported wheezing and fatigue levels. A random forest algorithm is then used to analyze the correlation strength between the logs and the patient's workload to determine triggering factors for symptom distress, including: Step 1: Starting from the critical timestamp, obtain the corresponding symptom log records, including information on the patient's self-reported wheezing and fatigue levels, and organize them into a structured dataset to obtain a preliminary set of symptom information. Step 2: For the initial set of symptom information, extract the records related to the load data, and use data matching technology to align the symptom manifestations with load fluctuations over time to determine the time-related dataset; Step 3: Based on the time-related dataset, the random forest algorithm is used to calculate the correlation strength between symptom manifestations and load fluctuations, analyze the interaction patterns between the two near the critical time stamp, and obtain the correlation strength results; ； Indicates the strength of the association between symptoms and workload. This represents the number of trees in a random forest. Indicates the first The reduction in Gini impurity in trees with symptom and load characteristics; Step 4: Based on the correlation strength results, if the correlation strength between a certain symptom and the load data exceeds a preset threshold, it is marked as a potential triggering factor, the corresponding symptom log fragment is obtained, and the key triggering factors are identified. Step 5: Based on the key triggering factors, extract the load data and symptom records for the relevant time periods, and analyze the changing trends of the two in different time windows through data comparison technology to determine the symptom triggering pattern; Step Six: For symptom triggering patterns, archive relevant information such as critical timestamps, load fluctuations, and symptom manifestations, and save them to the database using structured storage technology to obtain a complete symptom load profile.

7. The remote rehabilitation monitoring and intervention method based on a smart wearable device according to claim 1, characterized in that, The remote rehabilitation model parameters are updated based on symptom-related triggering factors. If the predicted participation rate decreases after the model parameters are adjusted, an intervention instruction sequence is generated to obtain an optimized intervention plan, including: Step 1: Obtain trigger factor data related to symptom distress from the database, classify and process this data, and organize the frequency of occurrence of trigger factors and corresponding symptom manifestations in chronological order to obtain a set of classified trigger factors; ； This indicates the frequency of occurrence of the k-th type of triggering factor. This indicates the total number of observation time points. Indicates the first Triggering factors at each point in time, Indicates an indicator function, when equal The value is 1 if the time factor is present and 0 otherwise. This formula is used to classify and process trigger factor data to obtain a frequency set. Step 2: Based on the categorized set of triggering factors, extract historical records related to remote rehabilitation, and use data matching technology to associate the triggering factors with the operating parameters of the rehabilitation model to determine the correspondence between the parameters and the symptoms. Step 3: Analyze the correlation between parameters and symptom distress, predict the trend of participation changes, and mark the participation as declining if the predicted participation value is lower than the preset threshold, and obtain the corresponding parameter adjustment requirements; Step 4: Based on the parameter adjustment requirements during the descent state, generate intervention instructions, and use serialization technology to arrange the intervention instructions according to priority to determine an ordered instruction sequence; Step 5: Based on the ordered sequence of instructions and the current state of the remote rehabilitation model, adjust the model parameters, generate the specific content of the optimized intervention, and obtain the adjusted intervention plan. Step Six: Based on the adjusted intervention plan, associate and store the intervention instruction sequence with the trigger factor data, save it to the database using structured storage technology, and determine the final basis for intervention execution; Step 7: Based on the final basis for intervention implementation, periodically extract the implementation records of the intervention plan from the database, analyze the changes in symptom distress through data comparison technology, and obtain an updated symptom management file.

8. The remote rehabilitation monitoring and intervention method based on a smart wearable device according to claim 1, characterized in that, The process involves optimizing the intervention plan and transmitting it to the patient's device in real time, adjusting the exercise intensity and duration, judging whether the patient's participation feedback data has improved, and obtaining the final rehabilitation effect evaluation index. This includes: Step 1: Obtaining the latest optimized plan data from the intervention plan database, using automated transmission technology, sending the plan content to the patient's device through a real-time transmission channel, and ensuring the integrity of the transmission process. Step 2: For the intervention plan transmitted to the patient's device, analyze the exercise intensity and duration parameters, and dynamically adjust the exercise task through the device's built-in control module to obtain the adjusted task execution instructions; Step 3: Based on the adjusted task execution instructions, record the feedback data of the patient device during the execution process, and use data acquisition technology to store the feedback data in real time to ensure a complete record of the feedback data; Step 4: For the complete record of feedback data, use data analysis tools to classify it. If the key indicators in the feedback data are lower than the preset threshold, they are marked as needing optimization, and the classified data analysis results are obtained. Step 5: Based on the data analysis results after classification, extract the feedback data of the state to be optimized, compare it with the effect evaluation criteria in the intervention plan, determine whether there is any deviation, and determine the specific deviation range; Step Six: For the specific deviation range, generate corresponding adjustment plan data, and use automated update technology to transmit the adjustment plan to the patient's device to obtain the updated exercise task configuration; Step 7: Based on the updated exercise task configuration, continuously monitor changes in the patient's device feedback data, track the trends using data analysis tools, and determine the final assessment criteria for rehabilitation effectiveness.

9. A remote rehabilitation monitoring and intervention system based on smart wearable devices, characterized in that, The system includes: The physiological signal acquisition and processing module is used to acquire physiological signals of patients during remote rehabilitation through wearable devices, including heart rate, respiratory rate and oxygen saturation data, and to use signal processing technology to filter out noise and obtain a clean physiological signal sequence. The feature extraction module is used to extract time-domain features, such as mean and standard deviation, and frequency-domain features, such as power spectral density, from the clean physiological signal sequence to determine the feature vector for subsequent analysis. The load status classification module is used to classify feature vectors using the support vector machine algorithm. If the classification result shows that the current physiological signal exceeds the preset normal range, it is judged as a potential increase in physical load and a load status label is obtained. The critical point determination module is used to calculate the load change rate by combining load status tags with historical motion data. If the change rate exceeds a preset threshold, the time when the critical point occurs is determined and the critical point timestamp is obtained. The symptom association analysis module is used to obtain symptom log data corresponding to the critical point timestamp, including the patient's self-reported wheezing and fatigue levels. The module uses a random forest algorithm to analyze the correlation strength between the logs and the workload to determine the triggering factors of symptom distress. The rehabilitation model update and intervention generation module is used to update the parameters of the remote rehabilitation model based on the triggering factors of symptom distress. If the predicted participation decreases after the model parameters are adjusted, an intervention instruction sequence is generated to obtain an optimized intervention plan. The intervention implementation and effect evaluation module is used to optimize the intervention plan and transmit it to the patient's device in real time, adjust the intensity and duration of exercise, determine whether the patient's participation feedback data has improved, and obtain the final rehabilitation effect evaluation indicators.

Images