A control method for functional rehabilitation equipment of affected limbs after breast cancer surgery

Abstract

The present invention is a control method for a functional rehabilitation device for affected limbs after breast cancer surgery. First, the physiological and activity data of the patient are collected by a variety of sensors integrated on the device, and key features are extracted from these data by applying signal processing technology; then, the extracted features are analyzed by using machine learning algorithms including support vector machines and random forests, and a personalized rehabilitation program is generated or adjusted based on this analysis, wherein the rehabilitation program is optimized according to the patient's progress, preferences and feedback; then, the performance of the patient in performing rehabilitation training is monitored in real time, and the intensity and complexity of the training are adjusted by using a dynamic difficulty adjustment algorithm, and the dynamic difficulty adjustment algorithm adopts a reinforcement learning optimization adjustment strategy; finally, cloud computing technology is used to synchronize and store the patient's rehabilitation data, and data mining and pattern recognition algorithms are used to analyze the rehabilitation progress, and based on this analysis, an optimization algorithm is applied to adjust the feedback to the patient and the rehabilitation therapist to optimize the rehabilitation strategy and improve the rehabilitation efficiency.

Description

Technical Field

[0001] The invention relates to a control method for rehabilitation equipment, in particular to a control method for functional rehabilitation equipment for affected limbs after breast cancer surgery. Background Art

[0002] Although the current control methods for postoperative limb functional rehabilitation equipment provide support for patients' recovery in many aspects, there are still a series of deficiencies and drawbacks, which limit the efficiency and effect of rehabilitation to a certain extent and also affect the patient's rehabilitation experience. First, the existing rehabilitation equipment often adopts a "one-size-fits-all" rehabilitation program, which lacks sufficient personalization and flexibility. This method does not fully take into account the individual differences in the patient's postoperative recovery status, such as age, gender, postoperative condition, and the patient's specific needs and preferences. Therefore, the rehabilitation plans provided by these devices may not be suitable for the specific situation of each patient, resulting in low rehabilitation efficiency and may even increase the patient's rehabilitation risk due to improper training intensity. Secondly, the existing rehabilitation control methods rarely monitor the patient's rehabilitation status in real time and dynamically adjust the rehabilitation plan accordingly. Although some high-end rehabilitation equipment can collect patients' rehabilitation data, the analysis and application of these data are usually not timely enough, resulting in the adjustment of the rehabilitation plan lagging behind the patient's actual recovery process. This delayed response not only prolongs the rehabilitation cycle, but also may miss the best time for rehabilitation and reduce the rehabilitation effect.

[0003] In addition, current rehabilitation equipment does not do enough in terms of psychological support. Postoperative patients are likely to encounter psychological challenges during rehabilitation, such as anxiety, depression, fear, and lack of motivation, and these changes in psychological state directly affect the rehabilitation effect. However, existing control methods rarely take into account the monitoring and intervention of psychological state, lack effective psychological support measures, and it is difficult to provide patients with a comprehensive rehabilitation program. In terms of technology integration, although some rehabilitation equipment has begun to try to integrate new technologies, such as virtual reality and artificial intelligence, to increase the attractiveness and interactivity of rehabilitation, the application of these technologies is often limited to specific links or functions, and lacks a systematic and comprehensive technology integration strategy. Therefore, these devices often fail to fully tap the potential of new technologies in rehabilitation and fail to provide a truly efficient, interactive and comprehensive rehabilitation environment.

[0004] At the same time, data privacy and security issues are also important drawbacks of existing rehabilitation equipment control methods. As rehabilitation equipment gradually develops towards intelligence and networking, a large amount of rehabilitation data is collected and transmitted. Without effective data protection measures, patients' privacy information is likely to be at risk of being leaked. At present, the protection measures for these data are not perfect enough to fully guarantee data security and patient privacy. Summary of the invention

[0005] The purpose of the present invention is to provide a control method for a functional rehabilitation device for an affected limb after breast cancer surgery, thereby solving some of the drawbacks and deficiencies pointed out in the background technology.

[0006] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: first, the physiological and activity data of the patient are collected by using a variety of sensors integrated on the device, and key features are extracted from these data by applying signal processing technology;

[0007] The extracted features are then analyzed using machine learning algorithms including support vector machines and random forests, and a personalized rehabilitation program is generated or adjusted based on the analysis, wherein the rehabilitation program is optimized based on the patient's progress, preferences, and feedback;

[0008] Then, by real-time monitoring of the patient's performance in performing rehabilitation training, a dynamic difficulty adjustment algorithm is used to adjust the intensity and complexity of the training, and the dynamic difficulty adjustment algorithm adopts a reinforcement learning optimization adjustment strategy; in addition, a learning algorithm based on user interaction is introduced to provide instant feedback and guidance in virtual reality to promote rehabilitation effects;

[0009] Finally, cloud computing technology is used to synchronize and store patient rehabilitation data, and data mining and pattern recognition algorithms are applied to analyze rehabilitation progress. Based on this analysis, optimization algorithms are used to adjust feedback to patients and rehabilitation therapists to optimize rehabilitation strategies and improve rehabilitation efficiency.

[0010] Furthermore, the integrated multimodal sensor system is specifically implemented by the following steps:

[0011] S1. Multimodal sensor integration: Using micro force sensor F, flexible strain gauge E, wearable heart rate monitor H and skin surface temperature sensor T, as well as optical sensor O and ultrasonic sensor U, the patient's rehabilitation data is comprehensively collected, where the data fusion function is:

[0012] D(t)=αF(t)+βE(t)+γH(t)+δT(t)+∈O(t)+ζH(t)

[0013] Among them,α,β,γ,δ,∈,ζ are weight coefficients adjusted according to the importance of different sensors, and t represents time;

[0014] S2. Advanced signal processing technology: Apply deep learning algorithm L and time series analysis S to process sensor data and extract features X(t), where: X(t) = L(D(t)) + S(D(t)), L and S represent the processing functions of deep learning and time series analysis, respectively, to identify specific rehabilitation activity patterns or abnormal physiological signals;

[0015] S3. Real-time data fusion and analysis: Implement the real-time data fusion framework R to analyze data through R(X(t)), where R stands for real-time analysis and feedback providing function, which aims to process and analyze data instantly and provide real-time feedback to patients and medical professionals;

[0016] S4. Application of adaptive algorithms: Develop an adaptive algorithm A to dynamically adjust the rehabilitation plan P according to the extracted features, where P = A(X(t)). A adjusts the rehabilitation plan based on the patient's current rehabilitation status and predicted rehabilitation trend to provide a personalized rehabilitation path.

[0017] Further, the machine learning algorithm first collects the patient's physiological and activity data D(t) using the device's integrated multimodal sensor system, including force, range of motion, heart rate, and skin temperature sensors;

[0018] Next, mathematical models are applied to the collected data for preprocessing and feature extraction:

[0019] P1. Data preprocessing: through function:

[0020] F pre (D(t)) = exp(-αt)·D(t) β

[0021] Where α and β are parameters adjusted according to the sensor type to standardize the data and reduce noise to ensure data consistency and comparability;

[0022] P2. Feature extraction and analysis: Using a combination of improved machine learning algorithms, including support vector machine (SVM) and random forest (RF), through the function:

[0023]

[0024] Where X represents the feature vector, ω1 and ω2 are algorithm weights, and γ is the adjustment parameter;

[0025] P3. Dynamic adjustment of rehabilitation program: According to the above analysis results, as well as the patient's progress p(t), preference h(t), and feedback f(t), the formula is dynamically adjusted:

[0026]

[0027] Generate or adjust the rehabilitation plan P, where P0 is the initial rehabilitation plan, κ, λ and μ are adjustment parameters, allowing real-time reflection of the patient's rehabilitation needs and feedback, ensuring the real-time and personalization of the rehabilitation plan.

[0028] Furthermore, the real-time monitoring of the patient's performance in performing rehabilitation training is achieved by:

[0029] S1. First, the multimodal sensor system integrated in the device is used to monitor the patient's performance in rehabilitation training in real time, including key indicators such as range of motion, force output and training completion, and the data is fed back to the control system of the rehabilitation device in real time;

[0030] S2, then using a reinforcement learning dynamic difficulty adjustment algorithm based on a deep Q network (DQN) or a policy gradient method to analyze and extract features from the real-time collected data, thereby learning the optimal training difficulty adjustment strategy in a continuous trial and error process, and adjusting the intensity and complexity of the training according to the patient's response to the current training difficulty;

[0031] S3. Generate a personalized rehabilitation training plan based on the output of the reinforcement learning model, making the rehabilitation process comfortable, efficient and personalized according to the comprehensive factors of the patient's physical recovery status, personal preferences and psychological state.

[0032] Furthermore, the multimodal sensor system utilizes a data processing algorithm:

[0033]

[0034] where D(t) represents the multimodal sensor dataset at time t, σ(·) is the data smoothing function, η is the weight parameter, and φ i (·) is the feature extraction function of the i-th sensor data, and n is the total number of sensors, which processes and analyzes the data collected from the sensors in real time;

[0035] Further included is an adaptive control system using a reinforcement learning based algorithm:

[0036]

[0037] Where a represents the system state, a represents a possible action, s′ represents the next state, P(s′|s,a) represents the probability of transferring from state s to state s′ by taking action a, R(s,a) is the immediate reward obtained by taking action a in state s, and γ is the discount rate. The parameters of rehabilitation training are adjusted according to real-time data feedback, including adjusting training intensity, modifying training mode, or providing personalized rehabilitation guidance.

[0038] Furthermore, the reinforcement learning dynamic difficulty adjustment algorithm:

[0039] First, through the integrated multimodal sensor system, the range of motion and key indicators of force output of the patient's affected limb are monitored in real time, using the function:

[0040]

[0041] To analyze the patient's motion data, where x irepresents the data collected by the i-th sensor, w i is the weight coefficient to identify and extract the features during the movement of the affected limb;

[0042] Next, we apply a reinforcement learning algorithm based on a Deep Q Network (DQN) or a policy gradient method by:

[0043]

[0044] Update strategy, where r represents the immediate reward, reflecting the patient's adaptability to the current training difficulty; s and s′ represent the current and next states, respectively, a and a′ represent the actions taken in these states, and γ is the discount rate of future rewards, so as to dynamically adjust the training difficulty and ensure that the training plan can achieve the rehabilitation effect without causing excessive pressure on the patient;

[0045] Finally, set the dynamic adjustment function:

[0046]

[0047] to adjust the training intensity, where I0 is the initial intensity, β is the adjustment sensitivity coefficient, and ΔQ(t′) represents the change in the Q value within the time period t′, reflecting the ability to dynamically adjust the training difficulty according to the patient's rehabilitation progress and feedback.

[0048] Furthermore, the personalized rehabilitation training program monitors the patient's physical recovery status including range of motion and force output of the affected limb and psychological feedback including stress indicators based on heart rate variability in real time through a multimodal sensor system integrated in the device, and fuses the data to construct a comprehensive recovery status model of the patient;

[0049] And use the function algorithm:

[0050]

[0051] where s t ,θ t and p t Respectively represent the physical state, behavioral data and psychological state at time t, ω represents the patient's personal preference, α and ξ are weight coefficients, and ψ are conversion functions, which process different types of data uniformly and provide personalized rehabilitation guidance;

[0052] Next, a dynamic difficulty adjustment algorithm based on reinforcement learning, including a deep Q-network (DQN) or a policy gradient method, is used, depending on the model:

[0053]

[0054] Conduct training, where r represents the immediate reward adjusted according to the patient's physical and psychological state, Δp represents the change in psychological state, which is used to punish the potential psychological burden caused by excessive training intensity, and γ and λ are the discount rate of future rewards and the weight coefficient of psychological burden, respectively;

[0055] Finally, expressed as a function:

[0056]

[0057] Optimize rehabilitation program, including P new and P current represent the new and current rehabilitation programs, respectively, κ is the adjustment factor, A gradient representing the adjustment of feedback based on the patient's state and behavior.

[0058] Furthermore, the user interaction learning algorithm:

[0059] A personalized rehabilitation environment created using virtual reality (VR) technology, combined with a reinforcement learning algorithm based on user interaction, dynamically adjusts the difficulty and complexity of rehabilitation training to suit the patient's physical recovery status, personal preferences, and psychological state;

[0060] Through comprehensive mathematical models:

[0061]

[0062] Implementation is performed, where s(t) represents the patient's physical state at time t, a(t) represents the action taken in this state, p(t) reflects the patient's psychological state or personal preference, α and β are adjustment coefficients used to balance the importance of immediate feedback and long-term rehabilitation goals, and f and g are conversion functions that convert physical state, action and psychological state into quantitative indicators of rehabilitation benefits;

[0063] Next, a reinforcement learning algorithm is used:

[0064]

[0065] Optimization is performed, where r(s,a) represents the immediate reward of action a in state s, γ is the discount rate, which represents the value of future rewards, s′ and a′ represent the subsequent state and action respectively, λ and μ are adjustment coefficients, and δp(t′) is the psychological state change function, which considers the impact of psychological burden on rehabilitation effect.

[0066] Furthermore, the method for optimizing rehabilitation strategy and improving rehabilitation efficiency adopts the following steps:

[0067] S1. Using cloud computing platform, real-time synchronization and storage of postoperative rehabilitation data D(t) of breast cancer patients, where:

[0068]

[0069] where d k (t) represents the kth rehabilitation data at time point t, λ k It is the decay parameter of data importance, which allows the weight of data to be adjusted dynamically over time;

[0070] S2. Use composite data mining and pattern recognition technology:

[0071]

[0072] where f k is the processing function for the k-th type of data, α k , β k,n and ω k Respectively represent the weight, periodic coefficient and frequency of the function, which are used to reveal key trends and patterns in the rehabilitation process;

[0073] S3. Based on the analysis results, the deep learning model is used to adjust the rehabilitation strategy:

[0074]

[0075] Where P prev It was the previous rehabilitation program. represents the loss function, taking into account the rehabilitation progress M and the previous plan P prev and the expected rehabilitation outcome R, η is the adjustment rate, It represents the gradient of the rehabilitation plan, which is used to guide the optimization direction of the plan and ensure the personalization and efficiency of rehabilitation training.

[0076] The present invention has multiple beneficial effects and aims to provide a comprehensive, personalized and efficient rehabilitation solution to maximize the restoration of patients' limb function and improve their quality of life. The following are the main beneficial effects of the control method:

[0077] 1. Personalized rehabilitation program: By utilizing cloud computing, data mining and pattern recognition technologies, combined with deep learning models, this method can dynamically adjust the difficulty and complexity of rehabilitation training according to the patient's physical recovery status, psychological state and personal preferences, providing each patient with a tailored rehabilitation program.

[0078] 2. Real-time progress monitoring and adjustment: Rehabilitation equipment can synchronize and store rehabilitation data in real time. Through continuous monitoring and analysis, key trends and patterns in the rehabilitation process can be discovered in a timely manner, so that the rehabilitation plan can be adjusted instantly according to the patient's progress, ensuring that rehabilitation training always remains in the most effective state.

[0079] 3. Reducing psychological burden: By considering changes in the patient's psychological state and incorporating them into the adjustment of the rehabilitation program, the present invention helps to reduce the psychological burden and pressure that the patient may encounter during the rehabilitation process, thereby promoting the patient's psychological well-being and creating a more comfortable and encouraging rehabilitation environment for the patient.

[0080] 4. Improve rehabilitation efficiency and effectiveness: Through precise data analysis and personalized training adjustments, the present invention can effectively improve the efficiency and effectiveness of rehabilitation training, help patients recover limb function faster, shorten the rehabilitation cycle, and thus improve the quality of life of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 The present invention is a flow chart of a control method for a functional rehabilitation device for an affected limb after breast cancer surgery.

[0082] Figure 2 The present invention is a flow chart of a method for real-time monitoring of a patient performing rehabilitation training. DETAILED DESCRIPTION

[0083] The specific implementation modes of the present invention are described in detail below with reference to the accompanying drawings.

[0084] A device control method for functional rehabilitation of affected limbs after breast cancer surgery involves first integrating multiple sensors, such as accelerometers, force sensors, electromyography (EMG) sensors, and angle sensors, on the rehabilitation device to comprehensively collect the patient's physiological and activity data during the rehabilitation process. These data cover key information such as the patient's range of motion, force output, and muscle activity. Subsequently, advanced signal processing techniques are applied, such as Fourier transform for analyzing frequency composition, wavelet transform for extracting features of non-stable signals, and machine learning algorithms such as random forests or support vector machines (SVM) to classify or identify these complex data, thereby extracting key features from the large amount of data collected. These features reflect the specific details of the patient's rehabilitation status, such as the accuracy of movement execution, the recovery of muscle strength, and the continuity of rehabilitation activities, providing rehabilitation therapists with accurate rehabilitation progress assessments, while also providing patients with immediate feedback to help them better understand their own rehabilitation status, optimize rehabilitation strategies, and improve rehabilitation efficiency.

[0085] Then, by applying machine learning algorithms, such as support vector machines (SVM) and random forests, the key features previously extracted from various sensor data are deeply analyzed, including but not limited to information such as the patient's range of motion, force output, and muscle activity. The role of machine learning algorithms in this process is to identify patterns and trends from a large amount of rehabilitation data, such as the improvement or possible challenges shown by patients in specific rehabilitation exercises. Such analysis helps add a layer of intelligent decision support in the generation or adjustment of rehabilitation programs. Based on this analysis, the system can generate or adjust a personalized rehabilitation program that not only takes into account the patient's current rehabilitation progress, but also fully considers the patient's personal preferences and real-time feedback, ensuring that rehabilitation training meets both the patient's physical rehabilitation needs and his or her psychological comfort. For example, if the analysis shows that the patient has good feedback on certain exercises and has made significant progress in these exercises, the system may increase the proportion of such exercises; conversely, if the patient has negative feedback on certain exercises or progresses slowly, the system may adjust accordingly, reduce the proportion of such exercises or find alternatives.

[0086] By monitoring the patient's performance in rehabilitation training in real time, the intensity and complexity of the training are adjusted using a dynamic difficulty adjustment algorithm, which is based on reinforcement learning to optimize the adjustment strategy. In this process, the sensors and monitoring systems in the rehabilitation equipment continuously collect data from the patient during rehabilitation training, such as the accuracy of movement execution, force output, and reaction time. These data are fed back to the control system of the rehabilitation equipment in real time, and the reinforcement learning algorithm evaluates the patient's current rehabilitation status and training performance based on these real-time data. The algorithm encourages patients to progress toward the target state through a preset reward mechanism, such as increasing the reward value to encourage patients to complete higher intensity or more complex training tasks. In this way, the dynamic difficulty adjustment algorithm can adaptively adjust the training difficulty and respond to the patient's rehabilitation progress and real-time performance in real time. If the patient shows good rehabilitation progress, the algorithm will increase the training difficulty, otherwise it will reduce it.

[0087] A user-interactive learning algorithm is introduced to provide instant feedback and guidance in a virtual reality (VR) environment to promote rehabilitation effects. This method creates an immersive rehabilitation training environment by integrating advanced VR technology and machine learning algorithms, allowing patients to perform rehabilitation exercises in a simulated three-dimensional space. The user-interactive learning algorithm can adjust the training content in real time according to the patient's behavior and reaction in the virtual environment, ensuring that the training plan is in line with the patient's current rehabilitation status and can stimulate the patient's active participation. Every movement and choice of the patient is captured and analyzed by the algorithm in real time. According to the patient's performance and feedback, the system adjusts the training difficulty, provides personalized guidance suggestions and positive incentives. This instant feedback mechanism not only helps patients better understand their rehabilitation progress and areas for improvement, but also makes the rehabilitation process more efficient and targeted.

[0088] Finally, it includes using cloud computing technology to synchronize and store patient rehabilitation data, and applying data mining and pattern recognition algorithms to analyze rehabilitation progress. Based on this analysis, optimization algorithms are used to adjust feedback to patients and rehabilitation therapists, thereby optimizing rehabilitation strategies and improving rehabilitation efficiency. Through the cloud computing platform, all data generated during the rehabilitation process, such as patients' movement data, physiological data, training feedback, and psychological state, can be uploaded, synchronized, and securely stored in real time to ensure data integrity and real-time performance. Subsequently, these rehabilitation data are deeply analyzed through data mining and pattern recognition algorithms to identify key progress and potential obstacles in the patient's rehabilitation process, such as discovering specific patterns in the rehabilitation process through cluster analysis, or predicting rehabilitation trends using time series analysis. Based on these analysis results, optimization algorithms, such as gradient descent or genetic algorithms, are applied to dynamically adjust rehabilitation training plans and strategies, as well as feedback for patients and rehabilitation therapists, to ensure that each step of training is as personalized and efficient as possible. In this way, rehabilitation therapists can adjust rehabilitation plans based on data-driven feedback provided by the system, and patients can also obtain immediate progress indicators and incentives to jointly promote the maximization of rehabilitation efficiency.

[0089] Embodiment 1:

[0090] Ms. Zhang, a postoperative breast cancer patient, is using a rehabilitation device that integrates a multimodal sensor system to help her restore the function of her affected upper limb.

[0091] S1: Multimodal sensor integration:

[0092] At the beginning of Ms. Zhang's rehabilitation program, the rehabilitation equipment collected data through the following sensors: a miniature force sensor F for measuring grip strength; a flexible strain gauge E for monitoring the range of motion of the joints; a wearable heart rate monitor H for recording heart rate changes during rehabilitation training; a skin surface temperature sensor T for monitoring local skin temperature to assess inflammation or blood circulation status; an optical sensor O for tracking and analyzing movement quality; and an ultrasonic sensor U for assessing soft tissue changes or muscle movement.

[0093] In order to reflect the importance of data from different sensors, the weight coefficients are set as follows:

[0094] α=0.2, β=0.2, γ=0.15, δ=0.15, ∈=0.15, ζ=0.15

[0095] These coefficients are designed to balance the contribution of various types of data in the assessment of rehabilitation progress.

[0096] On a typical rehabilitation training day, Ms. Zhang's device collected the following data (simplified example, actual values ​​may vary): F(t) = 30N, E(t) = 120°, H(t) = 85bpm, T(t) = 36.5°C, O(t) and U(t) data were converted to standardized values ​​that the device can analyze, set to 0.8 and 0.7 respectively (these values ​​are usually converted from the raw data of the sensor through a certain processing algorithm). Based on the above weight coefficients and data, the data fusion function is:

[0097] D(t)=0.2·30+0.2·120+0.15·85+0.15·36.5+0.15·0.8+0.15·0.7

[0098] =6+24+12.75+5.475+0.12+0.105=48.45

[0099] The fused data D(t) is 48.45, which will be used for further analysis to evaluate Ms. Zhang's rehabilitation progress and adjust the rehabilitation plan. Through this example, the embodiment can see that the integrated multimodal sensor system can comprehensively collect rehabilitation data from multiple dimensions and fuse these data into comprehensive indicators using preset weight coefficients. This method can not only comprehensively monitor the patient's rehabilitation status, but also provide an important basis for subsequent data analysis and rehabilitation strategy adjustment.

[0100] S2: Advanced signal processing technology:

[0101] Ms. Zhang's rehabilitation data has been collected and fused into a comprehensive index D(t) = 48.45 through an integrated multimodal sensor system. Now, in order to deeply understand this data and extract useful rehabilitation progress indicators and patterns, the rehabilitation device uses two advanced signal processing technologies: deep learning algorithm L and time series analysis S.

[0102] The deep learning algorithm L is set to focus on identifying complex rehabilitation movement patterns and patient behavior characteristics from the data. It processes and analyzes data through neural network models, such as convolutional neural networks (CNNs), which can capture subtle differences in rehabilitation movements and learn specific movement quality indicators from them. In Ms. Zhang's case, the deep learning algorithm was set to analyze the changes in D(t) over a time span and derive a quantitative indicator L(D(t)) = 25, which reflects a comprehensive assessment of Ms. Zhang's exercise execution quality and force output.

[0103] At the same time, time series analysis S is used to capture and predict the trend of rehabilitation data over time. This can be achieved through methods such as autoregressive moving average (ARMA) model or seasonal decomposition time series forecasting (STL). In this scenario, time series analysis may focus on the trend of heart rate changes or the periodic changes in skin surface temperature to evaluate the physiological impact of rehabilitation training. Setting time series analysis provides another indicator S(D(t)) = 10, which may represent the improvement of Ms. Zhang's heart rate recovery pattern and physiological state in rehabilitation activities.

[0104] Combining the outputs of these two technologies, the embodiment obtains the rehabilitation characteristic index:

[0105] X(t)=L(D(t))+S(D(t))=25+10=35

[0106] This indicator X(t) is 35, representing a comprehensive assessment of Ms. Zhang's rehabilitation training, including information on her exercise execution quality, force output, heart rate recovery, etc. By applying deep learning algorithms and advanced processing of rehabilitation data through time series analysis, the rehabilitation team can more accurately assess Ms. Zhang's rehabilitation progress, identify any possible barriers to rehabilitation, and discover potential health problems.

[0107] S3: Real-time data fusion and analysis

[0108] During Ms. Zhang's rehabilitation process, the rehabilitation equipment has collected rich data through an integrated multimodal sensor system, and processed this data using deep learning algorithms and time series analysis techniques to obtain a comprehensive rehabilitation feature index X(t) = 35. Now, in order to ensure that Ms. Zhang can get the most suitable training for her current stage of rehabilitation, the rehabilitation equipment uses a real-time data fusion and analysis framework R.

[0109] The specific implementation of setting R is completed through a real-time response system, which can instantly analyze the rehabilitation characteristic index X(t) and other relevant data and dynamically adjust the rehabilitation plan. In this framework, the embodiment sets a simplified response function to adjust the intensity and content of the feedback according to the real-time data. Considering the complexity of real-time analysis, the feedback adjustment function is set as:

[0110] R(X(t))=ξ·X(t)+ζ

[0111] Where ξ and ζ are adjustment coefficients, ranging from 0.5 to 1.5 and 5 to 15 respectively, to adapt to different rehabilitation needs and response speed requirements.

[0112] Taking Ms. Zhang as an example, during the training cycle, real-time analysis found that her rehabilitation characteristic indicators decreased, indicating that she may have encountered some difficulties in rehabilitation training. Assuming that ξ=1 and ζ=10 at this time, the real-time feedback adjustment function is R(35)=1·35+10=45. This result 45 indicates that Ms. Zhang's rehabilitation program needs to be appropriately adjusted, such as reducing the training intensity or introducing new rehabilitation activities to help her overcome the current challenges. Through this real-time data fusion and analysis framework, rehabilitation equipment can instantly identify problems and challenges in the rehabilitation process and respond by adjusting the rehabilitation program, thereby ensuring that the training plan always matches the patient's current rehabilitation status. This not only provides Ms. Zhang with the greatest support to help her recover more effectively, but also allows rehabilitation therapists to monitor the rehabilitation progress in real time and provide professional guidance based on the data and suggestions provided by the system.

[0113] S4: Application of adaptive algorithms

[0114] While Ms. Zhang was using the rehabilitation device with an integrated multimodal sensor system, her data was collected and analyzed in real time, generating a comprehensive rehabilitation characteristic index X(t)=35, and immediate adjustment suggestions R(X(t))=45 were obtained through the real-time data fusion and analysis framework. Now, in order to further optimize Ms. Zhang's rehabilitation path, an adaptive algorithm A is introduced, which aims to dynamically adjust the rehabilitation plan based on the patient's rehabilitation status and predicted rehabilitation trends.

[0115] The operating mechanism of the adaptive algorithm A includes analyzing the patient's rehabilitation progress, evaluating the effectiveness of the rehabilitation program, and making adjustments based on the patient's feedback. Specifically, the algorithm is implemented by the following formula: P = A (X (t)) = X (t) + ΔX, where ΔX represents the amount of adjustment based on patient feedback and predicted trends. In order to ensure the personalization and adaptability of the rehabilitation program, the calculation of ΔX is set based on a series of factors, including satisfaction with the rehabilitation progress, subjective feedback from the patient, and the degree of proximity to the long-term rehabilitation goals.

[0116] In Ms. Zhang's case, considering her fatigue response to some rehabilitation exercises and her satisfaction with the progress, ΔX = 5 was set, which means that the current rehabilitation program needs to be adjusted positively to add some low-intensity but targeted rehabilitation activities. Therefore, according to the adjustment of the adaptive algorithm, Ms. Zhang's new rehabilitation program score is P = 35 + 5 = 40.

[0117] This adjusted rehabilitation program score of P=40 will be used to reconfigure the settings of the rehabilitation equipment, introduce new rehabilitation activities, and adjust the intensity and duration of existing activities to better meet Ms. Zhang's actual rehabilitation needs and preferences. In this way, the adaptive algorithm enables the rehabilitation program to be flexibly adjusted based on the patient's real-time feedback and rehabilitation progress, ensuring that every step of training can provide the patient with maximum support and benefit.

[0118] Embodiment 2:

[0119] This paper discusses the application of machine learning algorithms in Ms. Zhang's rehabilitation process, especially the specific implementation of data preprocessing and feature extraction. The rehabilitation equipment used by Ms. Zhang collects her rehabilitation data D(t) in real time through an integrated multimodal sensor system, such as force sensor, range of motion sensor, heart rate monitor, and skin temperature sensor. In order to further analyze this data, it is first necessary to preprocess it to enhance the data quality and extract meaningful information.

[0120] P1: Data Preprocessing

[0121] In Ms. Zhang’s rehabilitation data processing process, a specific mathematical model was used in the data preprocessing step:

[0122] F pre (D(t)) = exp(-αt)·D(t) β

[0123] α and β are adjustment parameters, which are intended to standardize data and reduce noise. To ensure the adaptability and effectiveness of this process, for the strength sensor data, α=0.1 and β=2 are set in the embodiment, because the strength data may change rapidly over time and needs to be more sensitive to such changes; and for the heart rate data, considering its relative stability, α=0.05 and β=1 may be set in the embodiment.

[0124] Assuming that at a certain time point t=5 (unit: minutes), a set of rehabilitation data of Ms. Zhang includes a grip strength value of 30N and a heart rate of 85bpm. Applying the above data preprocessing formula, the embodiment processes these two data respectively:

[0125] For grip strength value (force sensor data), embodiments include:

[0126] F pre (30) = exp(-0.1×5)·30 2 =exp(-0.5)·900≈0.607×900=546

[0127] For heart rate (heart rate monitor data), the processing is:

[0128] Fpre (85) = exp(-0.05×5)·85 1 =exp(-0.25)·85≈0.779×85=66.215

[0129] Through this method, Ms. Zhang's original rehabilitation data became more standardized and clean after preprocessing, while retaining information that is critical to assessing rehabilitation progress. This step is the basis for subsequent deep learning models and time series analysis, ensuring the accuracy and effectiveness of data analysis. The preprocessed data can better reflect Ms. Zhang's actual rehabilitation status, providing a scientific basis for formulating or adjusting personalized rehabilitation plans, thereby ensuring the adaptability and efficiency of rehabilitation strategies.

[0130] P2: Feature extraction and analysis

[0131] After preprocessing, Ms. Zhang's rehabilitation data now needs further analysis to extract key features of rehabilitation progress. To this end, the rehabilitation device uses a combination of two machine learning algorithms: support vector machine (SVM) and random forest (RF). Both algorithms are very effective tools for processing complex data sets, identifying patterns and classification problems.

[0132] In order to integrate the advantages of these two algorithms and make the analysis results more accurate and reliable, the embodiment adopts the integrated model M ensemble (X). The algorithm weights ω1 = 0.6 and ω2 = 0.4 were set to reflect the contribution of the SVM and RF algorithms in the rehabilitation feature analysis, respectively. The adjustment parameter γ was set to 0.05 to control the influence of time decay and ensure that the model can adapt to the dynamic changes in the rehabilitation process.

[0133] At a specific time point, Ms. Zhang's feature vector X, after preprocessing and standardization, contains key indicators such as strength, range of motion, and heart rate. These features are analyzed by SVM and RF algorithms respectively, and the analysis results are set to SVM(X)=75 and RF(X)=65, which represent the rehabilitation progress evaluated from the perspective of SVM and RF algorithms respectively.

[0134] Substituting these values ​​into the integrated model, the embodiment obtains:

[0135]

[0136] Taking into account the integral The result when (t) is 5 minutes is about (19.8) (simplifying the calculation), we have:

[0137]

[0138] This result, namely Mensemble (X) = 3.41, providing a comprehensive rehabilitation feature score that combines the analysis of the SVM and RF algorithms and takes into account changes in rehabilitation progress over time. This score will be used to assess Ms. Zhang's current rehabilitation status and compared with previous scores to determine whether her rehabilitation is progressing as expected. Through this improved combination of machine learning algorithms, Ms. Zhang's rehabilitation team can obtain a more accurate and comprehensive rehabilitation progress analysis, which not only helps to better understand Ms. Zhang's rehabilitation needs, but also dynamically adjusts the rehabilitation plan based on the analysis results to ensure that rehabilitation training is both scientific and personalized.

[0139] P3: Dynamic adjustment of rehabilitation program

[0140] Now, the embodiment utilizes the dynamic adjustment formula:

[0141]

[0142] To generate or adjust Ms. Zhang's rehabilitation plan P. The initial rehabilitation plan P0 is designed to support Ms. Zhang's basic rehabilitation needs, taking into account her general state and expected progress in the early stage of rehabilitation. During the rehabilitation process, Ms. Zhang's progress p(t), preference h(t), and feedback f(t) will change in real time, which requires the parameters κ, λ, and μ in the formula to adjust the rehabilitation plan accordingly to reflect these changes.

[0143] The adjustment parameters were set to range from κ = 0.8, λ = 1.2, and μ = 0.5. These parameters were selected based on Ms. Zhang's actual performance and feedback in the early stage of rehabilitation, as well as the sensitivity to dynamic adjustment of the rehabilitation process.

[0144] It is assumed that within the first month of rehabilitation, Ms. Zhang's rehabilitation progress p(t) steadily increases, she shows a stronger preference h(t) for certain exercises, and her feedback f(t) shows that her satisfaction with the current rehabilitation program gradually increases. Specifically, the embodiment can be set that within the first 4 weeks of rehabilitation, her average rehabilitation progress score increases from 0.5 to 0.8, the preference changes little, remaining at around 0.7, and her satisfaction increases from the initial 0.6 to 0.9. Using these data and formulas for calculation, the embodiment can approximately estimate the value of adjusting the rehabilitation program.

[0145] If a simplified calculation is adopted, taking into account that the time t is in weeks, the embodiment obtains:

[0146] A(P,p(t),h(t),f(t))≈P0+4×0.8×(0.8+1.2×0.7+0.5×(0.9-0.6))

[0147] =P0+3.2×(1.54)=P0+4.928

[0148] This result means that according to Ms. Zhang's progress, preferences and feedback in the early stage of rehabilitation, her rehabilitation program needs to be adjusted accordingly to ensure that she can continue to rehabilitate in a supportive and motivating environment. This adjustment is not only based on objective rehabilitation progress assessment, but also takes into account the patient's personal preferences and psychological reactions, so as to more effectively promote the rehabilitation effect. Through this real-time and dynamic rehabilitation program adjustment method, Ms. Zhang's rehabilitation plan can continue to adapt to her specific needs and feedback, ensuring the personalization and efficiency of the rehabilitation process, and ultimately helping her to restore the function of the affected limb in the best way.

[0149] Embodiment 3:

[0150] How a patient named Ms. Zhang uses this device for rehabilitation training. Ms. Zhang started using this device to restore the function of her right upper limb after completing breast cancer surgery. The device monitors key indicators including range of motion, force output and training completion in real time through an integrated multimodal sensor system, and analyzes this data in real time through a specific data processing algorithm, providing real-time feedback to Ms. Zhang to optimize her rehabilitation strategy.

[0151] At a specific training stage, Ms. Zhang is performing a series of shoulder range of motion recovery exercises. The sensors of the device include force sensors, angle sensors, heart rate monitors, and skin temperature sensors, totaling n=4. During the training cycle, the data D(t) collected by these sensors include a force output value of 20N, a range of motion of 130 degrees, a heart rate of 85bpm, and a skin temperature of 36℃.

[0152] The specific implementation of the data processing algorithm is as follows:

[0153] The exponential smoothing function σ(D(t′)) is used to process real-time data, reduce noise and smooth short-term fluctuations. A uniform smoothing coefficient is set for all sensor data, that is, σ(x) = e -0.1x , to simplify the calculation.

[0154] For each sensor data, a different feature extraction function φ is used i (D(t i )). The weight parameter η is set to 0.25, indicating that all sensor types contribute equally to the final analysis results.

[0155] Substituting Ms. Zhang's sensor data into the formula, the embodiment performs the following calculation:

[0156] The data after smoothing is Since this is real-time data processing, the calculation process is simplified here and the original data is used directly.

[0157] The data after feature extraction is The results of each sensor data after a specific processing function are assumed to be: the power output characteristic value is 1.0, the motion range characteristic value is 1.3, the heart rate characteristic value is 0.85, and the skin temperature characteristic value is 0.36. Considering the weight parameter η comprehensively, the feature extraction result is 0.25×(1.0+1.3+0.85+0.36)=0.8775.

[0158] Finally, the rehabilitation training data analysis result P(D(t)) of Ms. Zhang will be combined with the smoothed data and feature extraction results. In order to simplify the display, the embodiment will add these two parts of the results to obtain a comprehensive analysis value to reflect Ms. Zhang's rehabilitation progress and effect. This analysis value will directly affect the dynamic adjustment of the training intensity and complexity of the rehabilitation equipment, ensuring that Ms. Zhang's rehabilitation training is in line with her actual recovery status and can be personalized according to her progress, preferences and feedback.

[0159] The device monitored her rehabilitation exercise performance in real time through an integrated multimodal sensor system, collecting data on key indicators including range of motion, force output, heart rate and skin temperature. Now, to further optimize Ms. Zhang's rehabilitation program, the device adopts an adaptive control system based on reinforcement learning.

[0160] Under the reinforcement learning framework, the device regards each of Ms. Zhang's rehabilitation activities as a decision-making process, during which the device dynamically adjusts the rehabilitation plan based on Ms. Zhang's real-time performance. Specifically, the algorithm uses the following formula to make decisions:

[0161]

[0162] Here, s represents the current state of recovery, a represents a possible action (such as adjusting the intensity of training or changing the type of training), s′ represents the new state that may be reached after taking the action, P(s′|s,a) is the state transition probability, and R(s,a) is the immediate reward obtained by taking a specific action. The discount rate γ is set to 0.9, reflecting the current value of future rewards.

[0163] Assume that at a certain moment, based on the analysis of Ms. Zhang's rehabilitation data, the device recognizes that adjusting the training intensity (action a1) will promote her rehabilitation efficiency. Through calculation, assuming that taking this action may bring an immediate reward R(s,a1)=5, it is predicted that Ms. Zhang's rehabilitation status has an 85% probability of improvement (P(s′|s,a1)=0.85), a 5% probability of remaining unchanged (P(s′|s,a1)=0.05), and a 10% probability of a slight decrease (P(s′|s,a1)=0.1). Substituting these values ​​into the above formula, the device calculates the expected value of taking action a1.

[0164] In this way, the device is able to dynamically adjust Ms. Zhang's rehabilitation program to maximize her rehabilitation benefits. For example, if the algorithm determines that adding a certain type of rehabilitation activity will bring Ms. Zhang greater immediate rewards and better rehabilitation results, the device will adjust her rehabilitation plan accordingly, increasing the frequency or intensity of this activity. Over time, the device continues to collect new data, re-evaluate and adjust the rehabilitation program to ensure that Ms. Zhang's rehabilitation path always matches her actual progress and personal preferences, making the rehabilitation process more efficient and personalized, and greatly increasing the success rate of rehabilitation.

[0165] Embodiment 4:

[0166] How a patient named Ms. Zhang used this device to carry out her rehabilitation journey. Ms. Zhang faced the challenge of upper limb functional rehabilitation after breast cancer surgery. Through its integrated multimodal sensor system, the device can monitor key indicators such as the range of motion and force output of Ms. Zhang's affected limb in real time, and adopts reinforcement learning algorithms including deep Q network (DQN) or policy gradient method to learn and adjust the training difficulty through a continuous trial and error process to maximize Ms. Zhang's rehabilitation progress.

[0167] During Ms. Zhang’s rehabilitation process, the device first used its multimodal sensor system to monitor and collect her rehabilitation training data in real time, including her range of motion and force output. By adopting the function:

[0168]

[0169] To analyze these data, x i represents the data collected from the i-th sensor (such as force sensor and angle sensor), and w i is the weight coefficient assigned to each sensor. When Ms. Zhang performs a specific exercise, the data collected from the force sensor is x1 = 15N, and the data collected from the angle sensor is x2 = 90°. The weight coefficients w1 = 0.6 and w2 = 0.4 are set to reflect the importance of different sensor data in rehabilitation training analysis.

[0170] Substituting the above data and weights into the analysis function, the embodiment obtains:

[0171]

[0172] This result, which is close to 1, shows that Ms. Zhang's performance at the current training difficulty is close to the ideal state, suggesting that the training difficulty may be moderate or slightly low. Therefore, based on the reinforcement learning algorithm, the device decides to appropriately increase the intensity or complexity of the training in the next training cycle to further promote Ms. Zhang's recovery. As the training progresses, the device continuously collects Ms. Zhang's performance data and continuously optimizes the training difficulty adjustment strategy through a deep Q network or a policy gradient method. This method enables the training plan to adapt to Ms. Zhang's recovery status and feedback in real time, ensuring that the training difficulty always remains at the optimal level that matches the challenge with her ability. Through this dynamic adjustment mechanism, Ms. Zhang is able to rehabilitate in an environment that supports her continued progress, ultimately achieving better rehabilitation results.

[0173] Ms. Zhang's rehabilitation device uses a reinforcement learning algorithm based on the deep Q network (DQN) or policy gradient method, which aims to optimize the difficulty of rehabilitation training through real-time data analysis. This method allows the device to adjust the rehabilitation training according to Ms. Zhang's immediate performance and feedback to ensure that the training is not only effective but also safe.

[0174] During Ms. Zhang’s rehabilitation, the device updated the training strategy using the following formula of the reinforcement learning algorithm:

[0175]

[0176] In this formula, r is the immediate reward, which reflects Zhang's adaptability to the current training difficulty. s and s' represent the current state and the next possible state, respectively, and a and a' represent the possible actions to be taken in these states. The discount rate γ of future rewards is set to 0.9 to evaluate the current value of future rewards.

[0177] In a specific training cycle, Ms. Zhang completed a specific rehabilitation activity, and the device gave an immediate reward r=5 based on her performance and feedback. The data monitored by the device showed that Ms. Zhang was able to successfully complete the training task and showed good adaptability to the current difficulty level. At this time, the device used the reinforcement learning algorithm to calculate the decision:

[0178] The current state s is that Ms. Zhang has successfully completed the designated rehabilitation exercises.

[0179] Possible action a is to increase the difficulty of training.

[0180] For the possible next state s', the maximum Q(s',a') value is set to 45, which represents the maximum reward that can be obtained by taking the best action in the future state. Based on this information, the device calculates the updated Q(s,a) value:

[0181] Q(s,a)=5+0.9×45=5+40.5=45.5

[0182] This calculation shows that increasing the difficulty of training is a favorable strategy because it leads to higher overall rewards.

[0183] Based on this updated Q value, the device decides to appropriately increase the difficulty of training in Ms. Zhang's next training cycle, such as increasing the number of training sessions or introducing more complex rehabilitation exercises. This decision is made to ensure that while Ms. Zhang continues to challenge herself, she can also maintain an appropriate rehabilitation progress to avoid overtraining and possible injuries. In this way, Ms. Zhang's rehabilitation training plan can be dynamically adjusted according to her rehabilitation status and instant feedback to ensure that she is always training at the most appropriate difficulty level.

[0184] Finally, as Ms. Zhang uses the rehabilitation device based on reinforcement learning algorithms, the device can not only monitor her training performance in real time, but also dynamically adjust the difficulty of the training according to her progress. To achieve this, the device uses a dynamic adjustment function:

[0185]

[0186] Where I0 is the initial intensity at the beginning of training, β is the adjustment sensitivity coefficient, which is used to determine the rate of adjustment of training intensity, and ΔQ(t′) represents the change in Q value within the time period t′, that is, the immediate reward change calculated by the reinforcement learning algorithm, which reflects Ms. Zhang’s rehabilitation progress and adaptability to training difficulty.

[0187] The initial intensity I0 of Ms. Zhang's rehabilitation training is set to 100 units. Taking into account the differences in adaptability and rehabilitation speed of different patients, the adjustment sensitivity coefficient β is set to 0.05, aiming to ensure that the adjustment of training intensity is neither too rapid to exceed the patient's adaptation range, nor too slow to affect the rehabilitation efficiency. During Ms. Zhang's specific training cycle (for example, the first four weeks), the Q value change ΔQ(t′) monitored by the device through the reinforcement learning algorithm shows that Ms. Zhang's adaptability to the current training difficulty is gradually increasing. During this time period, the average weekly Q value change ΔQ is set to approximately 2 units. Then, the embodiment can calculate the training intensity adjustment at the end of the four weeks:

[0188]

[0189] This means that after four weeks of training, the equipment adjusted the training intensity from the initial 100 units to 140 units based on Ms. Zhang's rehabilitation progress and her adaptability to the difficulty of the training. This adjustment is based on Ms. Zhang's actual feedback and rehabilitation progress, aiming to improve her rehabilitation efficiency while ensuring that the training plan does not put too much pressure on her. In this way, Ms. Zhang's rehabilitation equipment can dynamically adjust the training intensity in real time based on her feedback and rehabilitation progress, ensuring that the rehabilitation program always matches her actual needs and abilities. This personalized and adaptable rehabilitation training method can not only effectively promote Ms. Zhang's rehabilitation progress, but also greatly improve her rehabilitation experience and satisfaction.

[0190] Embodiment 5:

[0191] In Ms. Zhang's rehabilitation example, the embodiment entered the stage of using the output of the reinforcement learning model to generate a personalized rehabilitation training program. This program is to ensure that Ms. Zhang's rehabilitation process is not only efficient and personalized, but also to ensure that she feels comfortable during rehabilitation. This method takes into account Ms. Zhang's physical recovery status, personal preferences, and psychological state, and analyzes the data obtained by real-time monitoring through the multimodal sensor system integrated in the device.

[0192] Steps to implementing a personalized rehabilitation training program

[0193] 1. Multimodal sensor data collection: The device monitors Zhang’s range of motion and force output in real time to assess her physical recovery status; at the same time, it monitors her mental state through indicators such as heart rate variability, and collects data about her personal preferences.

[0194] 2. Data fusion and analysis: The following function algorithms are used to fuse and analyze the collected data:

[0195]

[0196] Among them, s t ,θ t and p t Represent the physical state, behavioral data and psychological state at time t, respectively, and ω represents Ms. Zhang's personal preference. Setting α = 0.1 and ξ = 1.2 as weight coefficients indicates the importance of different data types. Conversion function and ψ are used to transform these multidimensional data into a unified format that can be used for analysis.

[0197] Suppose at a certain time point, Ms. Zhang's physical condition score is s t = 0.8 (based on range of motion and force output), behavioral data score θ t =0.7 (based on personal participation in rehabilitation activities), psychological status score pt =0.9 (based on the heart rate variability index), and personal preference ω =0.85 (based on the preference for certain rehabilitation activities).

[0198] 3. Personalized rehabilitation guidance generation: Through the above-mentioned function algorithm, the device integrates and analyzes Ms. Zhang's rehabilitation data to generate a personalized rehabilitation training program. This program not only adjusts the training difficulty based on Ms. Zhang's physical recovery status and psychological feedback, but also takes into account her personal preferences, thus ensuring the personalization and targeting of the training plan.

[0199] After analyzing Ms. Zhang's rehabilitation data in this way, the device may find that she is well adapted to the current training and shows a stronger preference and better psychological response to some specific rehabilitation activities. Therefore, the device decides to increase the proportion of these activities and appropriately increase the difficulty of training to further promote Ms. Zhang's rehabilitation progress.

[0200] To further optimize her training plan, the rehabilitation device uses a dynamic difficulty adjustment algorithm based on reinforcement learning. This algorithm uses the deep Q network (DQN) or policy gradient method to dynamically adjust the training difficulty according to Ms. Zhang's physical and psychological state, ensuring that the training is effective without causing unnecessary psychological pressure on Ms. Zhang.

[0201] Dynamic Difficulty Adjustment Algorithm Application Example

[0202] In a certain training cycle, based on Ms. Zhang's physical recovery status and psychological feedback, the device uses the following model to make decisions:

[0203]

[0204] In this formula, r represents the immediate reward, which reflects Ms. Zhang’s adaptability after the training difficulty is adjusted; Δp represents the change in Ms. Zhang’s psychological state, which is used to identify and punish the psychological burden that may be caused by excessive training intensity; γ is set to 0.9, indicating the consideration of the current value of future rewards; λ is set to 0.1, indicating the weight of psychological burden in decision-making.

[0205] During one training session, the device observed that Ms. Zhang had a good adaptability to the current training difficulty and gave an immediate reward of r = 8. At the same time, by analyzing Ms. Zhang's heart rate variability and self-reported emotional state, the device estimated the change in psychological state Δp = -0.05, indicating that Ms. Zhang's psychological stress decreased slightly, which is a positive signal, indicating that the training intensity did not exceed her psychological tolerance.

[0206] Substituting these values ​​into the above decision model, the device calculates:

[0207]

[0208] Set prediction by DQN or policy gradient method, Then: Q(s,a)=8+0.9×10+0.005=17.005.

[0209] This calculation result shows that after considering Ms. Zhang's physical and psychological state, it is a reasonable choice to increase the difficulty of training appropriately, because it can bring Ms. Zhang a higher overall reward without adding too much psychological pressure. Through this dynamic difficulty adjustment algorithm based on reinforcement learning, Ms. Zhang's rehabilitation training program can flexibly adapt to her recovery progress and psychological state, ensuring that the training is both challenging and not excessive.

[0210] The rehabilitation equipment further optimized her rehabilitation program. This process utilized the function expression:

[0211]

[0212] Where P new and P current represent the new and current rehabilitation programs, respectively, κ is the adjustment factor, and It represents the gradient obtained according to Ms. Zhang’s state and behavior changes, that is, the adjustment of the optimal action value function Q in the decision-making process.

[0213] Steps to optimize your rehabilitation program:

[0214] Set Ms. Zhang's current rehabilitation plan P current Based on the results of previous evaluation and adjustment, the quantitative score was set to 80 points, representing a comprehensive rehabilitation exercise program including training intensity, frequency and type.

[0215] Select the adjustment factor κ, which determines the speed and magnitude of the adjustment of the rehabilitation program based on the feedback. Assume that κ = 0.05, which means that the adjustment of the rehabilitation program should reflect Ms. Zhang's immediate feedback while avoiding the potential negative impact caused by over-adjustment.

[0216] Based on Ms. Zhang’s state and behavior, the gradient is calculated through the reinforcement learning model Based on Ms. Zhang's performance and feedback after the most recent training cycle, the gradient calculation result is 2, which means there is a positive indication that the rehabilitation program should be adjusted in the direction of increasing difficulty.

[0217] Substituting these values ​​into the optimization equation, the new rehabilitation plan P is calculated. new :

[0218] P new =80+0.05·2=80+0.1=80.1

[0219] This calculation indicates that after taking into account the feedback from Ms. Zhang's most recent training session, her rehabilitation program was fine-tuned from 80 points to 80.1 points. Although this adjustment may seem very small, it is based on precise calculations and Ms. Zhang's real-time feedback, ensuring timely adjustments and personalized adaptation of the rehabilitation program. In this way, Ms. Zhang's rehabilitation program can continue to be fine-tuned according to her recovery progress, personal preferences, and psychological state. This dynamic adjustment mechanism ensures that the rehabilitation program always reflects Ms. Zhang's latest rehabilitation needs, thereby maximizing the rehabilitation effect while maintaining Ms. Zhang's comfort and satisfaction during the rehabilitation process. This process not only reflects the high adaptability of rehabilitation equipment, but also demonstrates the great potential of using advanced technology to provide patients with personalized rehabilitation programs.

[0220] Embodiment 6:

[0221] Continuing to explore Ms. Zhang's rehabilitation case, she is now using an advanced rehabilitation device that integrates virtual reality (VR) technology and user-interactive reinforcement learning algorithms. This device is designed to adapt to Ms. Zhang's physical recovery status, personal preferences, and psychological state through a personalized rehabilitation environment and dynamically adjusted training difficulty, thereby providing a highly personalized and effective rehabilitation experience.

[0222] Steps to implement the user interaction learning algorithm:

[0223] First, an immersive rehabilitation environment was created for Ms. Zhang based on VR technology. This environment was customized according to her personal preferences. For example, if Ms. Zhang prefers a natural environment, the rehabilitation scene could be a beautiful garden or a quiet forest. Next, a reinforcement learning algorithm based on user interaction was used to dynamically adjust the difficulty and complexity of the training. This process involves evaluating the effects of Ms. Zhang's actions in a specific physical state and making adjustments based on her psychological feedback.

[0224] Using a comprehensive mathematical model:

[0225]

[0226] To quantify the rehabilitation benefits. The adjustment coefficient α = 1 is set to indicate the importance of immediate feedback, while β = 0.1 is used to control the rate at which feedback decays over time. Functions f and g convert Ms. Zhang's physical state, actions taken, and psychological state into quantitative indicators of rehabilitation benefits.

[0227] In a typical rehabilitation session, Ms. Zhang's physical state (such as range of motion) is set to s(t) = 0.8, she chooses to perform a specific rehabilitation action a(t) = light stretching, and her psychological state (reflected by low stress level) is p(t) = 0.2. Through the above model, the device can calculate the contribution of these actions to the rehabilitation benefit and dynamically adjust the training difficulty accordingly.

[0228] Through this method, Ms. Zhang's rehabilitation plan can be adjusted in real time according to her recovery progress, personal preferences and psychological state, ensuring that her rehabilitation training in the VR environment is both effective and meets her personal needs. This process not only makes Ms. Zhang's rehabilitation more efficient and personalized, but also improves her satisfaction and participation in the rehabilitation process, helping to improve the overall rehabilitation effect.

[0229] To further optimize her rehabilitation training plan, the rehabilitation device uses a reinforcement learning algorithm, which specifically considers the impact of Ms. Zhang's physical recovery status, personal preferences, and psychological state, with the goal of dynamically adjusting the difficulty and complexity of the training to provide a personalized and efficient rehabilitation program.

[0230] Applications of Reinforcement Learning Algorithms:

[0231] 1. Instant reward (r(s,a)): When Ms. Zhang performs a specific rehabilitation activity (action a), the device evaluates her performance and gives her an instant reward of 5 points. This score reflects the effectiveness and adaptability of Ms. Zhang in completing the activity.

[0232] 2. Discount rate γ: The discount rate γ is set to 0.9, which means giving a higher current value to future rewards and emphasizing the importance of long-term rehabilitation effects.

[0233] 3. Adjustment coefficients λ and μ: In order to balance the impact of immediate rewards and psychological burden, set λ = 0.1 (the weight of psychological burden) and μ = 0.05 (the decay rate of psychological state change).

[0234] 4. Psychological state change function δpt′: It is assumed that within a period of time, Ms. Zhang’s psychological burden has increased, resulting in δp(t′) being 0.2 in a certain period of time, indicating that her psychological burden has increased relative to the beginning of training.

[0235] Optimization process of rehabilitation program:

[0236] Using the above parameters and variables, the device can calculate the Q value of Ms. Zhang taking a specific action in the current state. This Q value takes into account the value of immediate rewards, future rewards, and the impact of psychological burden. Set at a certain future state s′, the Q value of the best action a′ is expected to be 10. Then, the device will calculate the Q value of Ms. Zhang's current action:

[0237]

[0238] The impact of psychological burden is calculated integrally, and the overall impact value of psychological burden is set to 1 within a period of time Tt. In this way, the overall Q value is calculated as:

[0239] Q(s,a)=5+0.9×10-0.1×1=14-0.1=13.9

[0240] Results and implementation:

[0241] Based on the calculation results of the Q value, the rehabilitation device determines that the action taken in Ms. Zhang's current state is reasonable, but it is also necessary to consider reducing her psychological burden. Therefore, the device may decide to continue the current rehabilitation activities while introducing new strategies or activities to reduce psychological stress, such as adding more rest time or introducing psychological relaxation exercises.

[0242] Embodiment 7:

[0243] Through the rehabilitation journey of a woman named Ms. Zhang, this article explores how to use cloud computing platforms to synchronize and store rehabilitation data in real time, as well as specific steps to optimize rehabilitation strategies and improve rehabilitation efficiency.

[0244] Step 1: Real-time synchronization and storage of rehabilitation data

[0245] Ms. Zhang’s rehabilitation equipment uploads the collected rehabilitation data to the cloud computing platform in real time through the built-in communication module.

[0246] This platform not only provides huge storage space, but also enables real-time processing and analysis of data.

[0247]

[0248] Among them, d k (t) represents the kth type of rehabilitation data at time point t, which may include Ms. Zhang’s range of motion, power output, heart rate, etc. k is a decay parameter that reflects the importance of the k-th data and is used to adjust the weight of the data over time. This ensures that as the rehabilitation process progresses, more emphasis is placed on recent data rather than past data.

[0249] At a specific time point t, Ms. Zhang's rehabilitation data mainly includes two items: range of motion d1(t) = 120° and force output d2(t) = 15 N. The data importance attenuation parameter λ1 of the range of motion is set to 0.02, and the attenuation parameter λ2 of the force output is set to 0.01.

[0250] Substituting these values ​​into the formula to calculate D(t):

[0251] D(t)=e -0.02t 120+e -0.01t 15

[0252] If the embodiment considers a certain time point in the rehabilitation process, such as t=50 days, it is substituted into the above formula to calculate:

[0253] D(50)=e -0.02×50 120+e -0.01×50 15

[0254] The calculation results are:

[0255] D(50)=e -1 120+e -0.5 ·15≈0.368·120+0.607·15=44.16+9.105=53.265

[0256] The calculation result D(50) = 53.265 represents Ms. Zhang's comprehensive rehabilitation data score on the 50th day. In this way, the cloud computing platform can dynamically adjust the weight of the data to ensure timely adjustment and optimization of the rehabilitation strategy to adapt to Ms. Zhang's rehabilitation progress. Through this process, the rehabilitation team can monitor Ms. Zhang's rehabilitation data in real time, analyze her progress, and adjust the rehabilitation plan accordingly. For example, if it is found that the improvement rate of strength output is not as expected, exercises for strength training may be increased, or if it is found that Ms. Zhang's mental state has a downward trend, more psychological support may be introduced or the difficulty of training may be adjusted to avoid excessive stress.

[0257] In order to further optimize the strategy and improve rehabilitation efficiency, the rehabilitation team decided to use composite data mining and pattern recognition technology. This technology aims to deeply analyze Ms. Zhang's rehabilitation data, reveal key trends and patterns in the rehabilitation process, and provide her with more personalized and efficient rehabilitation guidance.

[0258] Step 2: Apply data mining and pattern recognition techniques:

[0259] For each type of rehabilitation data d k (t), select a specific data processing function f k For example, for range of motion data, you might use a function that reflects the rate of improvement; for strength output data, you might use a function that estimates the magnitude of the increase.

[0260] Calculation process formula:

[0261]

[0262] It integrates two main parts: one is the overall trend analysis of the data, and the other is the identification of periodic changes.

[0263] α k and β k,n are weight parameters, which are used to adjust f k (d k (t)) and the intensity of periodic changes.

[0264] ω k is the frequency at which each data type changes periodically.

[0265] Ms. Zhang's rehabilitation data is set to include two types: range of motion k=1 and strength output k=2, and the parameters are set as follows:

[0266] For the range of motion (k = 1), α1 = 0.8 is set, reflecting the greater importance of improving the range of motion in the rehabilitation process. In the periodic change analysis, β 1,1 =0.5, ω1=2π / 30, indicating that the improvement in the range of motion every 30 days shows an obvious cyclical change.

[0267] For power output (k = 2), α2 = 0.6 is set, reflecting that the increase in power output is also key to rehabilitation, but its importance is slightly lower than that of range of motion. 2,1 =0.3, ω2=2π / 45, which means that the periodic change of power output is 45 days. With this method, the rehabilitation team can comprehensively consider Ms. Zhang's rehabilitation data, not only analyze its overall trend, but also identify possible periodic changes, and provide Ms. Zhang with more refined rehabilitation guidance. For example, if it is found that Ms. Zhang's range of motion improvement increases significantly at the beginning of each month, then specific training can be added at this time point to promote greater rehabilitation benefits.

[0268] After completing the data mining and pattern recognition analysis, the rehabilitation team decided to use deep learning models to further optimize Ms. Zhang’s rehabilitation strategy. This step is to ensure that the rehabilitation program can more accurately adapt to Ms. Zhang’s current condition and drive her towards her desired rehabilitation goals.

[0269] Step 3: Deep learning model adjusts rehabilitation strategy:

[0270] Based on Ms. Zhang’s previous rehabilitation plan, the score or effect level was set at 80 points, which represents a comprehensive indicator of multiple dimensions including training frequency, intensity, duration, etc.

[0271] The loss function takes into account the current recovery progress M (based on the latest analysis results), the previous plan P prev, and the expected rehabilitation outcome R (e.g., the ultimate goal of Ms. Zhang’s rehabilitation is to return to a level of physical activity close to that before surgery). The purpose of the loss function is to quantify the deviation between the current plan and the expected outcome so that the plan can be adjusted to reduce this deviation.

[0272] The adjustment rate η is set to 0.1, which determines the magnitude of the plan adjustment. A smaller value of η means that the adjustment will be more cautious to avoid potential risks caused by over-adjustment. This step involves calculating the gradient of the loss function with respect to the rehabilitation plan to determine the optimal direction for adjusting the rehabilitation plan. This gradient indicates how to adjust the rehabilitation plan to most effectively reduce the deviation between the current plan and the expected results.

[0273] According to the deep learning model, the previous rehabilitation plan is combined with the gradient adjustment to obtain the optimized rehabilitation plan P n Phew. Use the formula:

[0274]

[0275] If the gradient of the loss function is set If the embodiment indicates that the intensity and frequency of training need to be increased, the new rehabilitation plan score may be adjusted to 82 points, which means that the optimization direction of the plan is to increase the intensity of rehabilitation to better suit Ms. Zhang's current recovery status and move towards her rehabilitation goals.

[0276] Through this method, Ms. Zhang's rehabilitation program is adjusted in real time based on her current progress and expected goals. This not only makes the rehabilitation process more personalized, but also ensures the efficiency and effectiveness of rehabilitation training, helping Ms. Zhang to restore her health and vitality in the best way. This process demonstrates the great potential of deep learning in the field of personalized medicine and rehabilitation, which can provide patients with customized rehabilitation programs to ensure the best rehabilitation results.

[0277] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A control method for a functional rehabilitation device for an affected limb after breast cancer surgery, characterized in that: By integrating a multimodal sensor system on the device to collect the patient's physiological and activity data, signal processing techniques are applied to extract key features from this data; Analyze the extracted features using machine learning algorithms including support vector machines and random forests to generate or adjust personalized rehabilitation plans that are optimized based on the patient's progress, preferences, and feedback; By real-time monitoring of the patient's performance in rehabilitation training, a dynamic difficulty adjustment algorithm is used to adjust the intensity and complexity of the training, and the dynamic difficulty adjustment algorithm adopts a reinforcement learning optimization adjustment strategy; a learning algorithm based on user interaction is introduced to provide instant feedback and guidance in virtual reality to promote rehabilitation effects; Use cloud computing technology to synchronize and store patient rehabilitation data, and apply data mining and pattern recognition algorithms to analyze rehabilitation progress, analyze and apply optimization algorithms to adjust feedback to patients and rehabilitation therapists, optimize rehabilitation strategies and improve rehabilitation efficiency; The integrated multimodal sensor system is specifically implemented by the following steps: S1. Multimodal sensor integration: Utilizing miniature force sensors ,Flexible strain gauge , Wearable Heart Rate Monitor and skin surface temperature sensor , and optical sensors and ultrasonic sensors , comprehensively collect the patient's rehabilitation data, where the data fusion function is: , in is the weight coefficient adjusted according to the importance of different sensors, Indicates time; S2. Advanced signal processing techniques: Applying deep learning algorithms and time series analysis Process sensor data and extract features ,in: , and The processing functions representing deep learning and time series analysis, respectively, identify rehabilitation activity patterns or abnormal physiological signals; S3. Real-time data fusion and analysis: Implementing a real-time data fusion framework ,pass Analyze the data, stands for Real-time Analysis and Feedback Provisioning Function, which processes and analyzes data and provides real-time feedback to patients and medical professionals; S4. Application of Adaptive Algorithms: Developing Adaptive Algorithms , dynamically adjust the rehabilitation plan based on the extracted features ,in , Adjust the rehabilitation plan based on the patient's current rehabilitation status and predicted rehabilitation trends, and provide a personalized rehabilitation path; The machine learning algorithm first collects the patient's physiological and activity data using the device's integrated multimodal sensor system, including force, range of motion, heart rate, and skin temperature sensors. ; Apply mathematical models to the collected data for preprocessing and feature extraction: P1. Data preprocessing: through function: , in and Parameters adjusted according to sensor type to standardize data and reduce noise, ensuring data consistency and comparability; P2. Feature extraction and analysis: Using an improved combination of machine learning algorithms, including support vector machine SVM and random forest RF, through the function: , in represents the feature vector, and is the algorithm weight, To adjust the parameters; P3. Dynamic adjustment of rehabilitation program: according to the results of P2 step analysis and the patient's progress , preference , and feedback , by dynamically adjusting the formula: , Generate or adjust rehabilitation plans ,in For the initial rehabilitation program, and To adjust parameters, it allows real-time reflection of patients’ rehabilitation needs and feedback, ensuring the real-time and personalized rehabilitation program; The user interaction learning algorithm: A personalized rehabilitation environment created using virtual reality (VR) technology, combined with a reinforcement learning algorithm based on user interaction, dynamically adjusts the difficulty and complexity of rehabilitation training to suit the patient's physical recovery status, personal preferences, and psychological state; Through comprehensive mathematical models: , To implement, Indicates that the patient is at time physical state, Represents the action taken in this state, Reflects the patient's psychological state or personal preference, and is a moderating factor that balances the importance of immediate feedback and long-term rehabilitation goals. and The conversion function converts physical state, action and psychological state into quantitative indicators of rehabilitation benefits; Using reinforcement learning algorithm: , Optimize, where Indicate action In Status Instant rewards for is the discount rate, representing the value of future rewards, and Represents the subsequent status and action respectively. and is the adjustment coefficient, It is a function of changes in mental state, and considers the impact of psychological burden on rehabilitation effect.

2. The control method for the functional rehabilitation equipment for the affected limb after breast cancer surgery according to claim 1 is characterized in that The method for implementing the real-time monitoring of the patient's performance in performing rehabilitation training: S1. Use the multimodal sensor system integrated in the device to monitor the patient's performance in rehabilitation training in real time, including key indicators such as range of motion, force output and training completion, and feed the data back to the control system of the rehabilitation device in real time; S2. Use a dynamic difficulty adjustment algorithm based on reinforcement learning, including deep Q-network DQN or policy gradient method, to analyze and extract features from real-time collected data, so as to learn the optimal training difficulty adjustment strategy in a continuous trial and error process, and adjust the intensity and complexity of training according to the patient's response to the current training difficulty; S3. Generate personalized rehabilitation training programs based on the output of the reinforcement learning model, making the rehabilitation process comfortable, efficient and personalized according to the comprehensive factors of the patient's physical recovery status, personal preferences and psychological state; The reinforcement learning dynamic difficulty adjustment algorithm: Through the integrated multimodal sensor system, the patient's range of motion and key indicators of force output are monitored in real time, using the function: , To analyze the patient's motion data, Representative The data collected by the sensors, is the weight coefficient to identify and extract the features during the movement of the affected limb; Apply reinforcement learning algorithms based on deep Q-networks (DQN) or policy gradient methods by: , Update strategy, where represents immediate reward, reflecting the patient's adaptability to the current training difficulty; and Represent the current and next states respectively. and Indicates the actions to be taken in these states, is the discount rate of future rewards to dynamically adjust the training difficulty; by setting the dynamic adjustment function: , To adjust the training intensity, is the initial strength, is the adjustment sensitivity coefficient, Indicates time period Changes in the internal Q value reflect the ability to dynamically adjust the difficulty of training based on the patient's recovery progress and feedback; The personalized rehabilitation training program, through the multimodal sensor system integrated in the device, monitors the physical recovery status of the patient's affected limb including range of motion and force output in real time, and psychological feedback including stress indicators based on heart rate variability, and fuses the data to build a comprehensive recovery status model for the patient; And use the function algorithm: , in and Respectively represent the time physical state, behavioral data, and psychological state, Indicate the patient's personal preference, is the weight coefficient, and It is a conversion function that processes different types of data in a unified manner and provides personalized rehabilitation guidance; Adopt a dynamic difficulty adjustment algorithm based on reinforcement learning, including deep Q network DQN or policy gradient method, according to the model: , Conduct training, including represents the immediate reward adjusted for the patient's physical and psychological state, Indicates the change in psychological state, used to punish the potential psychological burden caused by excessive training intensity. and They are the discount rate of future rewards and the weight coefficient of psychological burden; Expressed as a function: , Optimize rehabilitation program, including and Representing new and current rehabilitation programs, respectively. is the adjustment factor, A gradient representing the adjustment of feedback based on the patient's state and behavior.

3. The control method for the functional rehabilitation equipment for the affected limb after breast cancer surgery according to claim 2 is characterized in that The multimodal sensor system utilizes data processing algorithms to: , in Indicates time Multimodal sensor datasets at is the data smoothing function, is the weight parameter, For the feature extraction function for sensor data, and For the total number of sensors, real-time processing and analysis of data collected from sensors; Further included is an adaptive control system using a reinforcement learning based algorithm: , in Represents the system status, Represents action, Represents the next state, Indicates from the state Take Action Transfer to state The probability of To take action In Status The instant reward for To reduce the discount rate, adjust the parameters of rehabilitation training according to real-time data feedback, including adjusting training intensity, modifying training mode or providing personalized rehabilitation guidance.

4. The control method for the functional rehabilitation device for the affected limb after breast cancer surgery according to claim 1 is characterized in that The method for optimizing rehabilitation strategy and improving rehabilitation efficiency adopts the following steps: S1. Use cloud computing platform to synchronize and store the rehabilitation data of breast cancer patients in real time ,here: , in Represents at a point in time No. Rehabilitation data, It is the decay parameter of data importance, which allows the weight of data to be adjusted dynamically over time; S2. Use composite data mining and pattern recognition technology: , in For the The processing function of the class data, and Respectively represent the weight, periodic coefficient and frequency of the function, which are used to reveal key trends and patterns in the rehabilitation process; S3. Based on the analysis results, the deep learning model is used to adjust the rehabilitation strategy: , in It was the previous rehabilitation program. represents the loss function, taking into account the recovery progress , previous plans and expected rehabilitation outcomes is the adjustment rate, It represents the gradient of the rehabilitation plan, which is used to guide the optimization direction of the plan and ensure the personalization and efficiency of rehabilitation training.

Images