COPD patient rehabilitation training evaluation system based on multi-sensor information fusion

Abstract

The invention discloses a COPD patient rehabilitation training evaluation system based on multi-sensor information fusion. Belongs to the technical field of intelligent medical treatment, and particularly relates to the technical field of COPD patient rehabilitation training evaluation. The technical problems that an existing COPD patient rehabilitation training monitoring mode is insufficient in real-time performance and accuracy, the rehabilitation training result of the COPD patient cannot be evaluated in time, and therefore the rehabilitation training effect of the COPD patient is low easily are solved. The system can monitor the breathing parameters, the heart rate and the blood oxygen saturation of a patient in real time, an improved support vector regression model is adopted, data fusion and regression analysis are achieved, and the accuracy of lung function evaluation is improved. Through RSSI signal time sequence analysis, Lempel-Ziv complexity calculation, fuzzy discrete entropy and other feature extraction methods, deep fusion with heart rate and blood oxygen data is carried out, the learning ability of the model for dynamic changes of lung functions is improved, and prediction accuracy and stability are ensured.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent medical technology, and specifically relates to the field of rehabilitation training assessment technology for COPD patients. Background Art

[0002] In recent years, with the continuous improvement of people's living standards and the gradual enhancement of medical and health awareness, chronic obstructive pulmonary disease (COPD) has become an important health problem worldwide, and its prevention and treatment has received widespread attention. With the acceleration of urbanization, industrial development, traffic congestion and other factors, the air quality continues to decline, environmental pollution and long-term exposure to harmful substances such as dust and smoke have caused the incidence and risk of COPD to continue to rise. According to statistics, COPD has become the third leading cause of death in the world, and its high incidence and mortality rate have posed severe challenges to the public health system and the national economy.

[0003] COPD is not only characterized by its progressive, irreversible airflow limitation and chronic airway inflammation, but is also often accompanied by systemic lesions. Patients often experience frequent acute exacerbations during the course of the disease, requiring emergency medical treatment and additional drug intervention. This disease not only affects patients' daily life and work ability, but most patients also have other chronic diseases, such as cardiovascular disease and cancer, which further increases the complexity of disease management and the difficulty of treatment. In particular, patients with more severe airflow limitation are often faced with the direct threat of respiratory failure, and most COPD patients eventually die due to non-respiratory diseases.

[0004] In clinical treatment and rehabilitation management, pulmonary function rehabilitation and breathing training are widely considered to be important non-drug interventions to improve patients' quality of life and delay the progression of the disease. Traditional breathing training methods, such as pursed lip breathing, effective cough training, and respiratory muscle training, have been proven to effectively improve patients' respiratory muscle strength, improve lung ventilation function, and reduce hospitalizations and emergency department visits. However, different individuals may respond differently to different rehabilitation training movements. For example, for some patients, pursed lip breathing training can greatly improve their lung function status, while other patients may be more suitable for effective cough training.

[0005] Due to the deficiencies of existing rehabilitation equipment and monitoring systems in terms of real-time performance, accuracy and personalized guidance, it is impossible to timely evaluate the results of rehabilitation training for COPD patients, which can easily lead to low effectiveness of rehabilitation training for COPD patients, making it difficult for many COPD patients to obtain continuous and systematic rehabilitation training support. Summary of the invention

[0006] In order to solve the technical problem that the existing COPD patient rehabilitation training monitoring method has deficiencies in real-time and accuracy, and cannot timely evaluate the results of COPD patient rehabilitation training, which easily leads to low effectiveness of COPD patient rehabilitation training, the present invention provides a COPD patient rehabilitation training evaluation system based on multi-sensor information fusion, the system includes a data acquisition module, a data preprocessing module, a feature extraction module, a feature splicing module and a rehabilitation training effect prediction module; The data acquisition module is used to collect RSSI data and physiological index data of the test personnel during rehabilitation training; The data preprocessing module is used to perform denoising on RSSI data and perform missing compensation on physiological index data; The feature extraction module is used to extract features from RSSI data to obtain RSSI features, and to extract statistical features from physiological indicator data to obtain physiological features; The feature splicing module is used to perform standardization and feature fusion on the extracted RSSI features and physiological features to obtain a fused feature vector; The rehabilitation training effect prediction module is used to evaluate the lung function status of COPD patients by fusing feature vectors and judge the rehabilitation training effect; The rehabilitation training effect prediction module adopts an improved support vector regression model to evaluate the lung function status of COPD patients, and optimizes the hyperparameters of the support vector regression model through an improved particle swarm optimization algorithm to obtain an improved support vector regression model.

[0007] Furthermore, RFID tags and receiving devices are deployed in the data acquisition module. When the patient undergoes rehabilitation training, the RSSI value is recorded at fixed time intervals. The patient follows the guidance of the training video and completes each complete rehabilitation training action, which generates an RSSI data record corresponding to the action. At the same time, the physiological indicator monitoring device automatically generates physiological indicator data with a timestamp.

[0008] Furthermore, in the data preprocessing module, Gaussian filtering is performed on the RSSI data to remove noise, and the linear interpolation method is used to fill in the missing physiological indicator data.

[0009] Furthermore, the RSSI features include fuzzy discrete entropy, Lempel-Ziv complexity, kurtosis and skewness; the physiological features include sample size, mean, standard deviation, minimum value, 25% quantile, median, 75% quantile and maximum value.

[0010] Furthermore, the feature concatenation module performs Z-score normalization on RSSI features and physiological features by: conduct, It represents the feature after Z-score standardization. represents the original eigenvalue, is the mean of the original features in the training set, is the original characteristic standard deviation.

[0011] Furthermore, the hyperparameters of the support vector regression model are optimized by the improved particle swarm optimization algorithm as follows: Step 1: Initialize the parameters of the particle swarm optimization algorithm and the hyperparameters of the support vector regression model; on this basis, randomly generate the initial particle swarm and calculate the position and fitness value of each particle; Step 2: During the iterative optimization process, the speed and position of each particle are adjusted according to the update rules of the particle swarm optimization algorithm. The particle update is affected by the individual extreme value and the global optimal solution, so as to explore a better solution in the search space, calculate the updated fitness value, and compare the new and old fitness values ​​to decide whether to replace the current optimal solution; Step 3: After a certain number of iterations, the differential evolution algorithm is introduced to perform mutation and crossover operations on some individuals in the particle swarm. The differential evolution algorithm generates new candidate solutions through differential mutation and compares the fitness with the current solution. If the new candidate solution is better than the current solution, it is replaced; Step 4: After each iteration, check and update the individual optimal position and the global optimal fitness value; the individual optimal position is the best position found by the current particle during the iteration process, and the global optimal fitness value is the best solution among all particles; Step 5: Check whether the maximum number of iterations has been reached or the accuracy requirements have been met. If not, return to step 2 and continue to update the position and velocity of the particle. If the maximum number of iterations has been reached or a satisfactory optimal solution has been found, terminate the optimization process and output the final optimal fitness value and the corresponding hyperparameters.

[0012] Furthermore, the standard for determining the certain number of iterations in step 3 is 50 to 100 times.

[0013] The beneficial effects of the system of the present invention are: The patient's respiratory parameters, heart rate and blood oxygen saturation can be monitored in real time, and an improved support vector regression model is used to achieve data fusion and regression analysis, thereby improving the accuracy of lung function assessment. Compared with the traditional detection method that relies on large medical equipment, the present invention is based on a lightweight, non-invasive multi-sensor device, which enables patients to self-monitor at home or in a telemedicine environment, reducing their dependence on hospitals and professionals. Through feature extraction methods such as RSSI signal time series analysis, Lempel-Ziv complexity calculation, and fuzzy discrete entropy, the model is deeply integrated with heart rate and blood oxygen data to improve the learning ability of the model for dynamic changes in lung function and ensure the accuracy and stability of the prediction. In addition, medical staff can provide personalized rehabilitation training suggestions based on real-time monitoring data, dynamically adjust the training intensity, enhance patient compliance, and improve the rehabilitation effect. The optimized prediction model combines the particle swarm optimization algorithm to optimize the support vector regression model hyperparameters, improve the prediction accuracy and generalization ability, help accurately evaluate the lung function status, and assist doctors in formulating scientific rehabilitation plans. The system can alleviate the shortage of medical resources, improve the level of telemedicine and health management, and provide technical support for chronic disease management and public health services, thereby improving the quality of life of COPD patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a working flow chart of the system described in an embodiment of the present invention; Figure 2 Schematic diagram of RSSI data filtering in an embodiment of the present invention; Figure 3 A schematic diagram of a dual axis of physiological data in an embodiment of the present invention; Figure 4 This is a flowchart of the rehabilitation training effect prediction module in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of the present invention.

[0016] Embodiment 1, This embodiment provides a COPD patient rehabilitation training evaluation system based on multi-sensor information fusion. The system combines RFID wearable technology with a smart bracelet to achieve full-course training guidance and rehabilitation evaluation. The RFID smart sensing suit adopts a battery-free passive design to ensure comfort and convenience, while efficiently capturing respiratory movement signals; the smart bracelet continuously monitors key physiological parameters such as heart rate and blood oxygen, providing all-weather data support. Through multi-sensor data fusion and an improved support vector regression model, the system evaluates the standardization of training movements in real time, predicts rehabilitation progress, and improves training effects and patient participation.

[0017] The system includes a data acquisition module, a data preprocessing module, a feature extraction module, a feature splicing module and a rehabilitation training effect prediction module. The workflow diagram of the system is as follows: Figure 1 As shown: The data acquisition module is used to collect RSSI data and physiological index data of the test personnel during rehabilitation training; The data preprocessing module is used to perform denoising on RSSI data and perform missing compensation on physiological index data; The feature extraction module is used to extract features from RSSI data to obtain RSSI features, and to extract statistical features from physiological indicator data to obtain physiological features; The feature splicing module is used to perform standardization and feature fusion on the extracted RSSI features and physiological features to obtain a fused feature vector; The rehabilitation training effect prediction module is used to evaluate the lung function status of COPD patients by fusing feature vectors and judge the rehabilitation training effect; The rehabilitation training effect prediction module adopts an improved support vector regression model to evaluate the lung function status of COPD patients, and optimizes the parameters of the support vector regression model through a particle swarm optimization algorithm to obtain an improved support vector regression model.

[0018] Embodiment 2, This embodiment further limits the embodiment 1 and further illustrates the data acquisition module.

[0019] The RFID tags are bound to the surface of the human muscle group nodes to collect RSSI data of the breathing movements of the test personnel during rehabilitation training in the experimental environment; the smart bracelet collects physiological index information such as heart rate and blood oxygen during the synchronous rehabilitation training, such as Figure 3 This is a schematic diagram of the dual-axis physiological data collected by the smart bracelet, which is uploaded to the cloud platform for storage through the NB-IOT network. Researchers can use the cloud platform to obtain physiological indicator information data through computer terminals, and record the heart rate and blood oxygen data of the test subjects during rehabilitation training.

[0020] Taking pursed lip breathing training as an example, RFID tags and receiving devices are deployed in the experimental environment to ensure that the RSSI value can be recorded at fixed time intervals (every 10 seconds) when the patient performs pursed lip breathing training. The UHF RFID tag used is a passive liquid metal tag that complies with the ISO 18000-6C standard. The patient follows the guidance of the training video and generates an RSSI data record corresponding to each complete pursed lip breathing action. At the same time, the heart rate and blood oxygen monitoring equipment will automatically generate one or more physiological indicator data with timestamps.

[0021] The requirements of the pursed lip breathing training are: inhale for about 2 seconds - inhale slowly through the nasal cavity; exhale for about 6 seconds - slightly purse the lips and exhale slowly. In addition, in order to reduce the impact of individual differences, the conventional drug treatment regimen and nursing measures of all patients remained unchanged during the experiment; The passive liquid metal tag uses an Alien Higgs-3 chip, providing 96 bits of EPC storage space that can be expanded to 480 bits, as well as 512 bits of user storage space and a 64-bit unique TID. Its operating frequency range is 860 to 960 MHz, and the data transmission rate can reach 640 kbps.

[0022] Embodiment 3, This embodiment further limits the embodiment 1 and further illustrates the data preprocessing module.

[0023] The data preprocessing method is used to denoise the collected RSSI data and physiological information to ensure data quality. The RSSI filtering method uses Gaussian filtering. The RSSI data filtering diagram is shown in the figure. Figure 2 As shown in the figure, the missing physiological data are filled by linear interpolation. A linear relationship is constructed based on the existing data points to estimate the missing values ​​in a reasonable way, thereby maintaining the coherence and trend consistency of the data, retaining the dynamic change characteristics of the data, and avoiding the introduction of additional errors due to improper filling strategies.

[0024] The data preprocessing method is to perform Gaussian filtering on the original RSSI data, assign weights based on distance, better retain the RSSI change trend, and prevent the RSSI data from being interfered by environmental changes and multipath effects; the Gaussian filtering formula is as follows: ; in, is the index relative to the center point, is the standard deviation, is the RSSI value of the adjacent point, is the RSSI value after filtering, Indicates a specific point in time to be processed. Represents the range of Gaussian filtering, which is a constant.

[0025] Embodiment 4, This embodiment further limits the embodiment 1 and further illustrates the feature extraction module.

[0026] Based on the time series analysis method, the RSSI data under different rehabilitation training movements are feature extracted and a training sample set is constructed. Feature extraction methods include fuzzy discrete entropy, Lempel-Ziv complexity, kurtosis and skewness. Statistical feature extraction is performed on the physiological indicator information (heart rate and blood oxygen saturation) collected by the smart bracelet, including sample size, mean, standard deviation, minimum value, 25% quantile, median, 75% quantile and maximum value, to provide an overall overview of the data and understand its distribution and variability; Fuzzy discrete entropy can quantify the complexity of RSSI signals. Lempel-Ziv complexity can be used to measure the compressibility of signals, that is, the patterns and repeatability in the signals. Kurtosis and skewness reflect the sharpness of RSSI signals and the degree of deviation from the mean, respectively.

[0027] Embodiment 5, This embodiment further limits the embodiment 1 and further illustrates the feature splicing module.

[0028] The extracted RSSI features and physiological features are standardized to eliminate dimensional differences and feature fusion is performed. A unified feature vector is constructed by feature concatenation to enhance the data representation capability and provide high-quality input for the prediction model. Standardization processing, that is, Z-score standardization processing is performed on each feature. Since RSSI data is quite different from heart rate and blood oxygen data in terms of dimension and value range, direct concatenation may cause features with larger values ​​to dominate the model training process, thus affecting model performance. The Z-score standardization formula is as follows: ; in, represents the original eigenvalue, is the mean value of the feature in the training set, is the standard deviation of the feature. After normalization, each feature presents a standard normal distribution with a mean of 0 and a standard deviation of 1, which effectively eliminates the dimensional differences between different data sources and provides a unified scale for subsequent feature fusion.

[0029] The features extracted from each data source are directly spliced ​​before the model is input, and the original information of each data source is retained, which helps the model learn the potential nonlinear relationship between the features. The feature fusion formula is as follows: ; Among them, is the unified fusion feature vector. The RSSI feature vector after Z-score normalization is denoted as , and , the physiological feature vector after Z-score normalization is denoted as , and .

[0030] Example 6 This example further limits Example 1 and further explains the rehabilitation training effect prediction module. As Figure 4 shown is the workflow diagram of the rehabilitation training effect prediction module.

[0031] Use the improved particle swarm optimization algorithm to optimize the hyperparameters of the support vector regression model to improve the prediction performance of the model. Subsequently, the optimized support vector regression model is used to evaluate the pulmonary function status of COPD patients and judge the rehabilitation training effect.

[0032] An improved particle swarm optimization algorithm is introduced into the support vector regression model, and strategies of adaptive inertia weight and enhanced local search are added, making the optimization process more robust and able to converge to the global optimal or approximate optimal solution faster, further enhancing the fitting ability of the support vector regression model for complex relationships and improving the prediction accuracy.

[0033] Optimize the parameters of the support vector regression model using the improved particle swarm optimization algorithm. The specific steps are as follows: Step A: Initialize the key parameters of the algorithm, including the parameters of the particle swarm optimization (PSO) algorithm (such as learning rate, inertia weight, etc.) and the hyperparameters of the support vector regression model (SVR) (such as gamma). On this basis, randomly generate the initial particle swarm and calculate the position and fitness value of each particle to evaluate the performance of the current parameter combination on the objective function.

[0034] Step B: In the iterative optimization process, adjust the velocity and position of each particle according to the update rules of the particle swarm optimization algorithm. The update of the particle is affected by the personal extreme value (P best ) and the global optimal solution (G best ), so as to explore a better solution in the search space. Calculate the updated fitness value and compare the new and old fitness values to decide whether to replace the current optimal solution.

[0035] Step C: After a certain number of iterations, the differential evolution (DE) algorithm is introduced to perform mutation and crossover operations on some individuals in the particle swarm. The differential evolution algorithm generates new candidate solutions through differential mutation and compares the fitness with the original solution. If the new solution is better than the current solution, it is replaced. This mechanism can enhance the search capability, increase the probability of the algorithm jumping out of the local optimum, and make the overall optimization more stable and efficient.

[0036] Step D: After each iteration, check and update the individual optimal position and the global optimal fitness value. The individual optimal position is the best position found by the current particle during the iteration, while the global optimal fitness value is the best solution among all particles. This process continues to improve the efficiency of the search and the quality of the solution.

[0037] Step E: Check whether the maximum number of iterations has been reached or the accuracy requirements have been met. If not, return to step B and continue to update the position and velocity of the particle. If the maximum number of iterations has been reached or a satisfactory optimal solution has been found, terminate the optimization process and output the final optimal fitness value and the corresponding hyperparameters.

[0038] The hyperparameters optimized by the particle swarm optimization algorithm are used as input to train the support vector regression model and make predictions. On this basis, the model is evaluated to see whether it meets the preset accuracy requirements: if the model meets the accuracy requirements, the final evaluation results are output; if not, the parameters are readjusted and the training process is repeated until the performance indicators are met, and the algorithm process is finally terminated.

[0039] The trained model is deployed to the cloud to achieve remote prediction and monitoring functions. Researchers can access the cloud platform through computer terminals to obtain the lung function prediction results of COPD patients and adjust the rehabilitation training plan accordingly to improve the rehabilitation effect and personalized service level.

Claims

1. A COPD patient rehabilitation training evaluation system based on multi-sensor information fusion, characterized by: The system includes a data acquisition module, a data preprocessing module, a feature extraction module, a feature splicing module and a rehabilitation training effect prediction module; The data acquisition module is used to collect RSSI data and physiological index data of the test personnel during rehabilitation training; The data preprocessing module is used to perform denoising on RSSI data and perform missing compensation on physiological index data; The feature extraction module is used to extract features from RSSI data to obtain RSSI features, and to extract statistical features from physiological indicator data to obtain physiological features; The feature splicing module is used to perform standardization and feature fusion on the extracted RSSI features and physiological features to obtain a fused feature vector; The rehabilitation training effect prediction module is used to evaluate the lung function status of COPD patients by fusing feature vectors and judge the rehabilitation training effect; The rehabilitation training effect prediction module adopts an improved support vector regression model to evaluate the lung function status of COPD patients, and optimizes the hyperparameters of the support vector regression model through an improved particle swarm optimization algorithm to obtain an improved support vector regression model.

2. The COPD patient rehabilitation training evaluation system based on multi-sensor information fusion according to claim 1 is characterized in that: RFID tags and receiving devices are deployed in the data acquisition module. When patients undergo rehabilitation training, RSSI values ​​are recorded at fixed time intervals. Patients follow the guidance of the training video and generate an RSSI data record corresponding to each complete rehabilitation training action. At the same time, the physiological indicator monitoring device automatically generates physiological indicator data with timestamps.

3. The COPD patient rehabilitation training evaluation system based on multi-sensor information fusion according to claim 2 is characterized in that: In the data preprocessing module, Gaussian filtering is used to denoise the RSSI data, and linear interpolation is used to fill in the missing physiological indicator data.

4. The COPD patient rehabilitation training evaluation system based on multi-sensor information fusion according to claim 3 is characterized in that: The RSSI features include fuzzy discrete entropy, Lempel-Ziv complexity, kurtosis and skewness; the physiological features include sample size, mean, standard deviation, minimum value, 25% quantile, median, 75% quantile and maximum value.

5. The COPD patient rehabilitation training evaluation system based on multi-sensor information fusion according to claim 4 is characterized in that: The feature splicing module performs Z-score normalization on RSSI features and physiological features by: conduct, It represents the feature after Z-score standardization. represents the original eigenvalue, is the mean of the original features in the training set, is the original characteristic standard deviation.

6. The COPD patient rehabilitation training evaluation system based on multi-sensor information fusion according to claim 5 is characterized in that: The hyperparameters of the support vector regression model are optimized by the improved particle swarm optimization algorithm as follows: Step 1: Initialize the parameters of the particle swarm optimization algorithm and the hyperparameters of the support vector regression model; on this basis, randomly generate the initial particle swarm and calculate the position and fitness value of each particle; Step 2: During the iterative optimization process, the speed and position of each particle are adjusted according to the update rules of the particle swarm optimization algorithm. The particle update is affected by the individual extreme value and the global optimal solution, so as to explore a better solution in the search space, calculate the updated fitness value, and compare the new and old fitness values ​​to decide whether to replace the current optimal solution; Step 3: After a certain number of iterations, the differential evolution algorithm is introduced to perform mutation and crossover operations on some individuals in the particle swarm. The differential evolution algorithm generates new candidate solutions through differential mutation and compares the fitness with the current solution. If the new candidate solution is better than the current solution, it is replaced; Step 4: After each iteration, check and update the individual optimal position and the global optimal fitness value; the individual optimal position is the best position found by the current particle during the iteration process, and the global optimal fitness value is the best solution among all particles; Step 5: Check whether the maximum number of iterations has been reached or the accuracy requirements have been met. If not, return to step 2 and continue to update the position and velocity of the particle. If the maximum number of iterations has been reached or a satisfactory optimal solution has been found, terminate the optimization process and output the final optimal fitness value and the corresponding hyperparameters.

7. The COPD patient rehabilitation training evaluation system based on multi-sensor information fusion according to claim 6, characterized in that: The standard for determining the certain number of iterations in step 3 is 50 to 100 times.

Images