Rehabilitation path planning method, system, device and medium based on data analysis

Abstract

The application belongs to the technical field of data analysis, and provides a rehabilitation path planning method, system, device and medium based on data analysis, wherein the method comprises the following steps: collecting target data of a patient; performing decoupling analysis on fatigue characteristics of the patient according to the target data; constructing a multi-modal time series prediction model based on the target data and the decoupling analysis result, and predicting the residual tolerance time of the patient under the action of a central fatigue threshold or a peripheral fatigue threshold; and generating a dynamic scheduling adjustment instruction based on a first control instruction, a second control instruction and the residual tolerance time, so as to complete dynamic planning of a rehabilitation path of the patient. The application performs decoupling analysis on fatigue characteristics, extracts central state characteristics and peripheral state characteristics respectively, calculates corresponding fatigue indexes, and then outputs a control instruction through comparison between the central fatigue threshold and the peripheral fatigue threshold, thereby technically realizing clear distinction between central mental fatigue and peripheral muscle fatigue.

Description

Technical Field

[0001] This invention belongs to the field of data analysis technology, and specifically relates to a method, system, device and medium for rehabilitation pathway planning based on data analysis. Background Technology

[0002] Currently, stroke rehabilitation scheduling in hospital rehabilitation medicine departments typically uses a fixed timetable model, such as scheduling fixed physical therapy (PT) and occupational therapy (OT) sessions. This model has the following significant drawbacks: First, it cannot distinguish between different types of fatigue. Stroke patients often experience specific post-stroke fatigue, including mental fatigue of the central nervous system and muscle fatigue of the hemiplegic limbs. Existing monitoring methods usually rely solely on heart rate or blood pressure, which cannot distinguish whether the patient is experiencing attention exhaustion or muscle exhaustion. If a patient is mentally fatigued and forced to perform high-cognitive-load rehabilitation tasks, it will lead to decreased cooperation; if a patient is muscle fatigued and forced to train, it is very likely to induce compensatory movements or aggravate limb spasticity patterns. Second, it lacks preventive mechanisms against secondary injuries. The physiological tolerance of stroke patients fluctuates greatly, and existing monitoring technologies mostly rely on threshold alarms, i.e., they only prompt to stop after abnormal vital signs occur. This delayed feedback mechanism cannot predict the risk of impending spasticity or falls, making preventive intervention difficult. Third, the matching efficiency of rehabilitation resources is low. When patients experience a single type of fatigue (such as muscle fatigue only), existing treatments typically involve stopping all training altogether, resulting in a wasted precious treatment window and a lack of mechanisms for dynamically adjusting training content. Summary of the Invention

[0003] To address the problems in the background art, this invention proposes a rehabilitation pathway planning method, system, device, and medium based on data analysis.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: The data-driven dynamic programming method for rehabilitation pathways includes the following steps: Collect the patient's target data, which includes baseline data, central nervous system status data, and peripheral nervous system status data; Based on the target data, the patient's fatigue characteristics are decoupled and analyzed, and control commands are output. This includes: processing the target data to extract central and peripheral state characteristics; calculating the central fatigue index based on the central state characteristics and the peripheral fatigue index based on the peripheral state characteristics; preset the patient's peripheral fatigue threshold and dynamically adjust the patient's central fatigue threshold based on baseline data; comparing the real-time central fatigue index with the central fatigue threshold and the real-time peripheral fatigue index with the peripheral fatigue threshold, and outputting control commands. A multimodal time-series prediction model is constructed based on target data and decoupling analysis results to predict the remaining tolerance time of patients under the influence of central or peripheral fatigue thresholds. Based on control instructions and remaining tolerance time, dynamic scheduling adjustment instructions are generated to complete the dynamic planning of the patient's rehabilitation path.

[0005] Furthermore, baseline data included the patient's sleep duration and deep sleep percentage the previous night, as well as heart rate variability at rest in the morning.

[0006] Further, the target data is processed to extract central and peripheral state features, including the following steps: When extracting central state features, the frame difference method and optical flow method are used for eye movement video streams to calculate the variance of saccadic velocity and eyelid closure degree; fundamental frequency perturbation is extracted from speech signals; and real-time heart rate is extracted from photoplethysmography pulse waves. When extracting peripheral state features, time-frequency transformation is performed on the surface electromyography signal to calculate the median frequency of the power spectral density, and the slope is obtained by linear fitting of the sequence to characterize the spectral redshift trend; and jerk is calculated on the inertial sensor data to characterize the smoothness of the movement.

[0007] Furthermore, the central fatigue index is calculated based on the characteristics of the central nervous system state, including the following steps: A pre-trained logistic regression model is used, with peripheral state features as the input vector; The input vector is fed into the Sigmoid function to calculate the probability that the patient is in a state of mental fatigue; Multiplying the probability by 100 maps it to a scalar of 0-100, and the scalar is used as the central fatigue index.

[0008] Furthermore, the peripheral fatigue index satisfies: ; In the formula, I p This indicates the peripheral fatigue index; α and β The weighting coefficients are calculated based on information entropy to reflect the differences in sensitivity of different patients to changes in electromyography or movement; Norm represents normalization; SlopeMDF represents the median frequency slope; and Jerk represents the smoothness of movement.

[0009] Furthermore, a preset peripheral fatigue threshold for the patient is established, and the central fatigue threshold is dynamically adjusted based on baseline data, including the following steps: Peripheral fatigue values ​​are set based on the patient's limb function rating; Based on baseline data, the previous night's sleep quality score was selected and normalized to the 0-1 range, and the deviation of the ratio of low-frequency to high-frequency morning heart rate variability was selected and normalized to the 0-1 range. The normalized sleep quality score and deviation were used as correction factors. The correction factor is substituted into the linear bias formula to dynamically adjust the central fatigue threshold.

[0010] Furthermore, the linear bias formula is: Th c_current =Th c_default -[k1×(1-S sleep )+k2×D hrv ]; In the formula, Th c_current This represents the dynamically adjusted central fatigue threshold; Th c_default This represents the default alarm threshold for central nervous system fatigue; S sleep Indicates the previous night's sleep quality score; D hrv This indicates the deviation of the low-frequency to high-frequency ratio of morning heart rate variability; k1 and k2 are preset sensitivity coefficients.

[0011] Furthermore, the real-time central fatigue index is compared with the central fatigue threshold, and the real-time peripheral fatigue index is compared with the peripheral fatigue threshold, and a control command is output, including the following steps: If the central fatigue index is less than or equal to the central fatigue threshold and the peripheral fatigue index is less than or equal to the peripheral fatigue threshold, it is determined to be a suitable training state, and the current rehabilitation schedule is maintained. If the central fatigue index is greater than the central fatigue threshold and the peripheral fatigue index is less than or equal to the peripheral fatigue threshold, it is determined that central fatigue is dominant, and the first control command is output to trigger the circuit breaker mechanism for cognitive tasks. If the central fatigue index is less than or equal to the central fatigue threshold and the peripheral fatigue index is greater than the peripheral fatigue threshold, it is determined that peripheral fatigue is dominant, and the second control command is output to trigger the pause or resistance unloading of the resistance training module. If the central fatigue index is greater than the central fatigue threshold and the peripheral fatigue index is greater than the peripheral fatigue threshold, it is determined to be high-level combined fatigue, and all active training tasks are forcibly suspended.

[0012] Furthermore, a multimodal time-series prediction model is constructed based on the target data and decoupling analysis results to predict the patient's remaining tolerance time under the influence of the central or peripheral fatigue threshold, including the following steps: The model input layer is constructed by: building a dynamic input layer using central fatigue features and peripheral fatigue features to obtain a high-dimensional time series tensor; and building a static stream input layer using baseline data to obtain the daily baseline vector. Design a baseline-gated LSTM unit, including: mapping the baseline vector of the static stream input layer to a weight factor through a fully connected layer, and applying the weight factor to the forget gate activation function of the LSTM unit; The model output layer is constructed and connected to a regression analyzer. Based on the processing results of the baseline-gated LSTM unit, the remaining tolerance time is output in a single numerical form. The remaining tolerance time is used to predict the time it will take for the patient to reach the set peripheral fatigue threshold or the dynamically adjusted central fatigue threshold under the current training load.

[0013] Furthermore, based on control instructions and remaining tolerance time, dynamic scheduling adjustment instructions are generated to complete the dynamic planning of the patient's rehabilitation pathway, including the following steps: If central fatigue is determined based on the first control command, tasks with cognitive loads higher than the set cognitive load are suspended, and tasks with cognitive loads lower than the set cognitive load are executed. If peripheral fatigue is determined based on the second control command, resistance training of the patient's limbs is suspended and replaced with speech therapy, cognitive training or training of the unaffected limbs. Based on the remaining endurance time, the fatigue threshold is predicted, and based on the fatigue threshold, continuous training sessions are broken down into intermittent training sessions with rest periods inserted.

[0014] A data-driven dynamic programming method system for rehabilitation pathways includes: The acquisition unit is used to collect the patient's target data, which includes baseline data, central nervous system status data, and peripheral status data. The analysis unit, used to perform decoupled analysis of patient fatigue characteristics based on target data, includes: The normalization module is used to process the target data and extract central and peripheral state features. The calculation module is used to calculate the central fatigue index based on the central state characteristics and the peripheral fatigue index based on the peripheral state characteristics. The adjustment module is used to preset the patient's peripheral fatigue threshold and dynamically adjust the patient's central fatigue threshold based on baseline data. The output module is used to compare the real-time central fatigue index with the central fatigue threshold and the real-time peripheral fatigue index with the peripheral fatigue threshold, and output control commands. The prediction unit is used to construct a multimodal time-series prediction model based on target data and decoupling analysis results to predict the patient's remaining tolerance time under the influence of central or peripheral fatigue thresholds. The planning unit is used to generate dynamic scheduling adjustment instructions based on control commands and remaining tolerance time, thereby completing the dynamic planning of the patient's rehabilitation path.

[0015] Furthermore, the prediction unit includes: The first input construction module is used to construct the model input layer, including: constructing a dynamic input layer through central fatigue features and peripheral fatigue features to obtain a high-dimensional time series tensor; and constructing a static stream input layer through baseline data to obtain the daily baseline vector. The second feature fusion and extraction module is used to design the baseline-gated LSTM unit, including: mapping the baseline vector of the static flow input layer to a weight factor through a fully connected layer, and applying the weight factor to the forget gate activation function of the LSTM unit to output a temporal feature hidden vector that incorporates individual baseline weights; The third time prediction output module is used to construct the model output layer. Its input is the time-series feature hidden vector output by the second module. By connecting to the regression analyzer, based on the processing results of the baseline-gated LSTM unit, it outputs the remaining tolerance time in a single numerical form. The remaining tolerance time is used to predict the time when the patient will reach the set peripheral fatigue threshold or the dynamically adjusted central fatigue threshold under the current training load.

[0016] An electronic device, comprising: Memory, used to store computer programs; The processor implements the above method when executing computer programs stored in memory.

[0017] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0018] The beneficial effects of this invention are: 1. This invention performs decoupled analysis on fatigue characteristics, extracts central state characteristics and peripheral state characteristics respectively and calculates the corresponding fatigue index, and then outputs control commands by comparing the central fatigue threshold (dynamically adjusted based on baseline data) and the peripheral fatigue threshold, thereby technically realizing a clear distinction between central mental fatigue and peripheral muscle fatigue.

[0019] This solution addresses the shortcomings of existing technologies that rely solely on single indicators such as heart rate and blood pressure to differentiate fatigue types. It avoids the decline in patient cooperation caused by conducting high cognitive load training when mentally fatigued, as well as the problem of compensatory movements or worsening of limb spasms caused by forced training when muscles are fatigued, thus improving the targeting and safety of rehabilitation training. 2. This invention constructs a multimodal temporal prediction model specifically to predict the remaining tolerance time of patients when they reach the central or peripheral fatigue threshold, breaking through the lag limitation of the existing "threshold alarm" technology; the prediction results can predict the fatigue state that patients will experience in advance, effectively avoiding secondary injuries such as spasms and falls induced by large fluctuations in physiological tolerance, and providing a more reliable rehabilitation safety guarantee for stroke patients and other people with unstable physiological states. 3. When a patient experiences only a single type of fatigue, this invention can switch to an appropriate task (such as speech therapy, cognitive training, etc.) by adjusting instructions, thus avoiding the waste of the treatment window. At the same time, it breaks down long training sessions and inserts rest periods based on fatigue prediction results, achieving flexible adaptation of rehabilitation tasks. This solution maximizes the use of limited rehabilitation resources, improves the continuity and overall efficiency of rehabilitation training, and solves the problem of low resource matching efficiency under the existing fixed scheduling model.

[0020] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description and the drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart of the data analysis-based dynamic planning method for rehabilitation pathways of the present invention is shown; Figure 2 A structural diagram of the data analysis-based dynamic planning system for rehabilitation pathways of the present invention is shown. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] like Figure 1 The diagram shows a data-driven dynamic programming method for rehabilitation pathways, which includes the following steps: S1: Collect the patient's target data, which includes baseline data, central nervous system status data, and peripheral nervous system status data.

[0025] S2: Decouple the patient's fatigue characteristics based on the target data and finally output control commands, including: S201: Process the target data and extract central and peripheral state characteristics; S202: Calculate the central fatigue index based on the central state characteristics and the peripheral fatigue index based on the peripheral state characteristics; S203: Preset the patient's peripheral fatigue threshold and dynamically adjust the patient's central fatigue threshold based on baseline data; S204: Compare the real-time central fatigue index with the central fatigue threshold and the real-time peripheral fatigue index with the peripheral fatigue threshold, and output control commands.

[0026] S3: Construct a multimodal time-series prediction model based on target data and decoupling analysis results to predict the remaining tolerance time of patients under the influence of central or peripheral fatigue thresholds.

[0027] S4: Based on control instructions and remaining tolerance time, generate dynamic scheduling adjustment instructions to complete the dynamic planning of the patient's rehabilitation path.

[0028] S1 is used to acquire multidimensional signals from patients. Three sets of data are required to address the characteristics of stroke patients. The first set is baseline data: This involves reading the patient's sleep duration and deep sleep percentage from the previous night via an interface, as well as heart rate variability (HRV) at rest in the morning. The second set is central nervous system data: This uses a non-contact visual sensor to capture the patient's eyelid closure, blink frequency, and saccade speed; simultaneously, a microphone is used to collect the patient's fundamental frequency jitter and speech pauses to characterize focus and mental state; additionally, a photoplethysmography (PPG) sensor worn on the patient's earlobe or wrist is used to acquire pulse wave signals in real time for subsequent extraction of real-time heart rate sequences to characterize the patient's mental workload and autonomic nervous system regulation. The third set is peripheral nervous system data: This uses surface electromyography (EMG) sensors attached to key muscle groups on the affected limb (such as the tibialis anterior and biceps brachii) to acquire the median frequency and average power frequency of EMG signals; or, in conjunction with an inertial sensor, to acquire the smoothness of the affected limb's movement trajectory.

[0029] It should be noted that other methods can be used for signal acquisition. For example, portable electroencephalography (EEG) devices can be used to directly monitor characteristic wavebands of brain waves (such as the ratio of alpha waves to beta waves) to replace eye movement and speech analysis. In peripheral fatigue detection, for scenarios using rehabilitation robots, the changes in current and torque data fed back from the motor can be used to estimate the patient's muscle strength decline, replacing surface electromyography (EMG) patches.

[0030] S2 is used for decoupling analysis of fatigue characteristics. It establishes a dual-channel signal processing logic to map the three sets of heterogeneous data collected in step one into two normalized quantitative indicators (central load index and peripheral load index).

[0031] In S201, when extracting central state features, the frame difference method and optical flow method are used for eye-tracking video streams to calculate the variance of saccadic velocity and the percentage of eyelid closure over the pulley over time; fundamental frequency jitter is extracted from the speech signal; simultaneously, real-time heart rate is extracted from the photoplethysmography as a basic physiological feature; in addition, when extracting peripheral state features, time-frequency transformation is performed on surface electromyography (sEMG) signals to calculate the median frequency (MDF) of the power spectral density, and the slope is obtained by linear fitting of the MDF sequence to characterize the spectral redshift trend; jerk is calculated for inertial sensor data to characterize the smoothness of movement.

[0032] In S202, two normalized fatigue indices are calculated using both a probabilistic model and a physical model, as detailed below: 1) Central fatigue index I c The calculation process: A pre-trained logistic regression model is used to process the above eye-tracking features X. eye Speech features X voice and real-time heart rate X hr As the input vector X.

[0033] The input vector is fed into the Sigmoid function to calculate the probability P that the patient is in a state of "mental fatigue". Finally, the probability P is multiplied by 100 to map to a scalar I between 0 and 100. c The logistic regression model mentioned above was pre-trained under supervised supervision based on a large amount of historical data from stroke patients. The labels used during training were derived from the clinically common Stanford Sleepiness Scale (SSS) or the Rating of Perceived Exertion Scale (RPE).

[0034] The Sigmoid function is: (1) In the formula, P represents the probability that the patient is in a state of mental fatigue, and X represents the input vector, which includes eye-tracking features X. eye Speech features X voice and real-time heart rate X hrw and b are the parameters of the logistic regression model, which are obtained by pre-training based on a large amount of historical data from stroke patients.

[0035] 2) Peripheral fatigue index (I p The calculation process is as follows: A linear weighted model based on the entropy weight method is adopted. Since muscle fatigue has significant physical characteristics, the system directly fuses electromyographic redshift characteristics and movement tremor characteristics. The specific formula is as follows: (2) In the formula, I p This indicates the peripheral fatigue index; α and β The weighting coefficients are calculated based on information entropy to reflect the differences in sensitivity of different patients to changes in electromyography or movement; Norm indicates normalization. and It is a core-peripheral fatigue characteristic; among which The slope represents the median frequency. The more negative the slope (the larger the absolute value), the faster the MDF decreases, which means that the peripheral fatigue process, such as changes in muscle fiber recruitment patterns and accumulation of metabolic products, is developing more severely. Jerk value indicates the smoothness of movement. Under fatigue, muscle control decreases, and movements become more unstable and tremors increase, leading to a significant increase in Jerk value. An increased Jerk value reflects poor muscle contraction coordination and unstable force output, which is a direct manifestation of decreased motor control due to peripheral fatigue.

[0036] In S203, the peripheral fatigue value Th is first set based on the patient's limb function rating. p Then, the initial settings are made based on the patient's limb function rating (Brunnstrom classification) upon admission (the worse the function, the lower the threshold), and remain unchanged in a single training session or be manually adjusted by the doctor as the rehabilitation cycle progresses.

[0037] Th, the central fatigue threshold c Calibration is usually required. The calibration procedure involves using the baseline data (sleep quality score S1) collected in S1. sleep Using the HRV baseline as a correction factor, a daily physiological state correction coefficient is constructed to personalize the default general alarm threshold. Specifically, the previous night's sleep quality score (S...) is selected. sleep The values ​​are normalized to 0-1, and the deviation D of the morning HRV low-frequency to high-frequency ratio is selected. hrv It is then normalized to 0-1, and these two variables between 0 and 1 are used as two key correction factors. Among them, S... sleep The lower the value, the less rest the central nervous system tolerates; D hrvThe higher the level, the more likely it is to indicate an imbalance in sympathetic / parasympathetic nervous system regulation and poor stress recovery ability.

[0038] The correction factor is substituted into the linear bias formula to dynamically adjust the central fatigue threshold. The linear bias formula is as follows: Th c_current =Th c_default -[k1×(1-S sleep )+k2×D hrv ];(3) In the formula, Th c_current This represents the dynamically adjusted central fatigue threshold; Th c_default This represents the default alarm threshold for central nervous system fatigue; S sleep Indicates the previous night's sleep quality score; D hrv This indicates the deviation of the low-frequency to high-frequency ratio of morning heart rate variability; k1 and k2 are preset sensitivity coefficients.

[0039] It should be noted that the central fatigue threshold Th c Defined as the critical value that triggers the mental fatigue circuit breaker. The system presets a general benchmark value (e.g., Th). c_default =75), this value is derived from the average logistic regression probability of a general stroke population reaching "severe drowsiness" on the Stanford Drowsiness Scale. Peripheral fatigue threshold Th p Defined as the threshold that triggers muscle protection mechanisms. The system is initialized based on the patient's limb motor function rating (e.g., Brunnstrom stage) upon admission. For example, Brunnstrom stage I-II patients have extremely poor muscle tolerance, and the system sets their Thr... p Initialize to a lower value (e.g., 60); while patients in stages IV-V are initialized to a higher value (e.g., 80).

[0040] It should be further noted that in S203, the default alarm threshold for system central fatigue is set to Th. c_default (For example, 75 points). The system uses a linear bias formula to calculate the personalized threshold for the day. For example, suppose a patient had insomnia last night (S...). sleep =0.4) and morning HRV showed poor autonomic regulation (D hrv =0.6), according to the formula, its alarm threshold will be significantly reduced from the default 75 points to around 63 points. This means that as long as the patient shows even a slight slowing of eye movements or a decrease in attention on the same day, the system will trigger the "central fatigue circuit breaker" in advance, thereby achieving a more stringent protection mechanism.

[0041] In S204, the real-time calculated index is compared with the dynamic threshold, and a control signal is generated through logic gate circuits or software logic: if the central fatigue index ≤ central fatigue threshold and the peripheral fatigue index ≤ peripheral fatigue threshold, it is determined to be a suitable training state, and the system maintains the current rehabilitation schedule unchanged; if the central fatigue index > central fatigue threshold and the peripheral fatigue index ≤ peripheral fatigue threshold, it is determined to be central fatigue dominant, and the first control command is output to trigger the circuit breaker mechanism of cognitive tasks; if the central fatigue index ≤ central fatigue threshold and the peripheral fatigue index > peripheral fatigue threshold, it is determined to be peripheral fatigue dominant, and the second control command is output to trigger the suspension or resistance unloading of the resistance training module; if the central fatigue index > central fatigue threshold and the peripheral fatigue index > peripheral fatigue threshold, it is determined to be high complex fatigue, and all active training tasks are forcibly suspended.

[0042] It should be noted that, for prediction models, in addition to the deep learning algorithms mentioned above, traditional machine learning algorithms such as Support Vector Machine (SVM) or Random Forest can also be used for classification prediction. Logical judgment methods based on expert rule bases can also be employed, such as setting multi-level warning thresholds. Although the accuracy is slightly lower, the computational load is small, and it is easy to implement on low-computing-power devices.

[0043] S3 constructs a baseline-gated multimodal temporal prediction model that can address the issue of significant differences in physiological tolerance among different patients. The specific steps are as follows: S301: Construct the model input layer, including: constructing a dynamic input layer using central fatigue features and peripheral fatigue features to obtain a high-dimensional time series tensor; and constructing a static stream input layer using baseline data to obtain the daily baseline vector.

[0044] Specifically, the input layer of the S301 model adopts a dual-stream structure. The dynamic stream uses a sliding window mechanism to extract the real-time central and peripheral state feature sequences (including real-time heart rate, electromyography MDF slope, eye movement PERCLOS value, etc.) extracted in step two, forming a high-dimensional time series tensor X. t The static stream directly inputs the baseline vector Sbase (including the duration of deep sleep the previous night, morning HRV value, etc.) obtained in step one. This baseline vector remains fixed during a single model training process and is from the same data source as the baseline data used to dynamically adjust the central fatigue threshold.

[0045] S302: Designing a baseline gated long-term memory network (LSTM, Long Short Memory Network) The Term Memory (TSM) LSTM unit includes a fully connected layer that maps the baseline vector of the static flow input layer to weight factors. These weight factors are then applied to the forgetting gate activation function of the LSTM unit. This allows the model to adaptively adjust the feature retention weights based on individual patient baseline differences (such as sleep quality and baseline physiological state). It also specifically strengthens weak fatigue-related features identified by S2 (such as declining electromyography (EMG) trends and abnormal heart rate fluctuations), improving the fatigue prediction adaptability for patients with different physiological tolerance levels. For example, for patients with poor sleep, the forgetting gate parameters are automatically adjusted, giving the model a higher retention weight for recent weak EMG decline trends, thus predicting the fatigue threshold more quickly.

[0046] S303: Construct the model output layer and connect it to the regression analyzer. Based on the processing results of the baseline-gated LSTM unit, output the remaining tolerance time in a single numerical form. The remaining tolerance time is used to predict the time it will take for the patient to reach the set peripheral fatigue threshold or the dynamically adjusted central fatigue threshold under the current training load.

[0047] For example, S303 connects to a regression analyzer at the model output layer, and outputs the remaining endurance time T in a single numerical form based on the fused features processed by S302 (including individual baseline weights and real-time fatigue trend features). rem The remaining tolerance time T rem Define the preset peripheral fatigue threshold Th p Or dynamically adjusted central fatigue threshold Tr c It accurately predicts the specific time when a patient will reach any fatigue threshold under the current training load. For example, the system calculates and outputs information in real time: "Under the current load, the affected lower limb is expected to reach the peripheral fatigue threshold Th in 12.5 minutes." p The prediction result will be used as a feedforward signal and directly input into S4 for adaptive scheduling, which will trigger task reorganization before fatigue actually occurs.

[0048] Based on the above judgment results, S4 generates a dynamic scheduling adjustment instruction, including the following steps: S401: If central fatigue is determined based on the first control command, then tasks with cognitive loads higher than the set cognitive load (such as dual-task walking, fine grasp training) are suspended, and tasks with cognitive loads lower than the set cognitive load (such as functional electrical stimulation therapy, passive joint movement) are performed.

[0049] S402: If peripheral fatigue is determined based on the second control command, the patient's resistance training is suspended and replaced with speech therapy, cognitive training or training of the unaffected limbs to give the affected muscles time to recover and avoid inducing hypertonic spasm. S403: The remaining endurance time can predict the approaching fatigue threshold. Based on the fatigue threshold, continuous (long-term) training can be broken down into intermittent training with rest periods.

[0050] It should be noted that S4 primarily describes adjusting rehabilitation schedules. Alternatively, the system could communicate directly with rehabilitation training equipment (such as exoskeleton robots or electrical stimulators) to directly adjust the equipment's output parameters (such as the level of assistance and stimulation frequency), rather than simply providing scheduling suggestions.

[0051] like Figure 2 The diagram shows a data analysis-based dynamic programming method system for rehabilitation pathways, which includes: The acquisition unit is used to collect the patient's target data, which includes baseline data, central nervous system status data, and peripheral status data. The analysis unit, used to perform decoupled analysis of patient fatigue characteristics based on target data, includes: The normalization module is used to process the target data and extract central and peripheral state features. The calculation module is used to calculate the central fatigue index based on the central state characteristics and the peripheral fatigue index based on the peripheral state characteristics. The adjustment module is used to preset the patient's peripheral fatigue threshold and dynamically adjust the patient's central fatigue threshold based on baseline data. The output module is used to compare the central fatigue index with the central fatigue threshold and output the first control command, and to compare the real-time peripheral fatigue index with the peripheral fatigue threshold and output the second control command. The prediction unit is used to construct a multimodal time-series prediction model based on target data and decoupling analysis results to predict the patient's remaining tolerance time under the influence of central or peripheral fatigue thresholds. The planning unit is used to generate dynamic scheduling adjustment instructions based on the first control instruction, the second control instruction, and the remaining tolerance time, thereby completing the dynamic planning of the patient's rehabilitation path.

[0052] Furthermore, the prediction unit includes: The first input construction module is used to construct the model input layer, including: constructing a dynamic input layer through central fatigue features and peripheral fatigue features to obtain a high-dimensional time series tensor; and constructing a static stream input layer through baseline data to obtain the daily baseline vector. The second feature fusion and extraction module is used to design the baseline-gated LSTM unit, including: mapping the baseline vector of the static flow input layer to a weight factor through a fully connected layer, and applying the weight factor to the forget gate activation function of the LSTM unit to output a temporal feature hidden vector that incorporates individual baseline weights; The third time prediction output module is used to construct the model output layer. Its input is the time-series feature hidden vector output by the second module. By connecting to the regression analyzer, based on the processing results of the baseline-gated LSTM unit, it outputs the remaining tolerance time in a single numerical form. The remaining tolerance time is used to predict the time when the patient will reach the set peripheral fatigue threshold or the dynamically adjusted central fatigue threshold under the current training load.

[0053] In summary, compared with existing technologies that rely solely on heart rate thresholds for passive alarms, this invention, based on simulation test datasets and comparative experimental verification, has the following advantages: First, it significantly reduces the response delay in fatigue detection: traditional methods rely solely on heart rate threshold alarms. Due to the physiological lag in heart rate response (cardiac drift), the heart rate typically doesn't significantly exceed the threshold until 3-5 minutes after muscle exhaustion, by which time cramps have often already occurred. This system directly captures the precursors of nerve and muscle fatigue by monitoring electromyographic redshift (MDF slope) and eye movement micro-expressions. Simulation tests show that the average detection time of the fatigue threshold using this invention is 180 to 240 seconds earlier than that of a simple heart rate monitoring system.

[0054] Secondly, it significantly reduces the false alarm rate in rehabilitation scheduling. Existing methods are easily affected by patient emotions. For example, if a patient's heart rate increases due to tension, the existing system may misjudge this as "physical fatigue" and force a shutdown, resulting in the incorrect interruption of effective training time. This invention introduces a "central-peripheral" dual-flow decoupling mechanism, enabling the system to accurately distinguish between "high heart rate caused by tension" and "high heart rate caused by exhaustion." In simulation tests including noise interference, the false fatigue alarm rate of this system is reduced from 35% in existing technologies to less than 8%, ensuring the continuity of rehabilitation training.

[0055] Third, it increases the density of rehabilitation tasks per unit of time. Existing technologies typically adopt a "complete rest" strategy after detecting fatigue, leading to idle rehabilitation resources. This invention employs a dynamic task reorganization strategy (i.e., switching between cognitive and motor tasks). Experimental data show that within the same 60-minute rehabilitation session, the patient's effective active participation time increased from the traditional 35 minutes to 48 minutes, and the utilization rate of rehabilitation resources increased by approximately 37%.

[0056] An electronic device, characterized in that it comprises: Memory, used to store computer programs; When the processor executes the computer program stored in memory, it implements the above-mentioned data analysis-based dynamic programming method for rehabilitation pathways.

[0057] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the aforementioned data analysis-based dynamic programming method for rehabilitation pathways. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data-driven dynamic programming method for rehabilitation pathways, characterized in that, Includes the following steps: Collect the patient’s target data, which includes baseline data, central nervous system status data, and peripheral nervous system status data; The process involves decoupling and analyzing the patient's fatigue characteristics based on target data, and outputting control commands. This includes: processing the target data to extract central and peripheral state characteristics; calculating a central fatigue index based on the central state characteristics and a peripheral fatigue index based on the peripheral state characteristics; presetting a peripheral fatigue threshold for the patient and dynamically adjusting the central fatigue threshold based on baseline data; comparing the real-time central fatigue index with the central fatigue threshold and comparing the real-time peripheral fatigue index with the peripheral fatigue threshold, and outputting control commands. A multimodal time-series prediction model is constructed based on target data and decoupling analysis results to predict the remaining tolerance time of patients under the influence of central or peripheral fatigue thresholds. Based on control instructions and remaining tolerance time, dynamic scheduling adjustment instructions are generated to complete the dynamic planning of the patient's rehabilitation path.

2. The data analysis-based dynamic planning method for rehabilitation pathways according to claim 1, characterized in that, The baseline data consisted of the patient's sleep duration and deep sleep percentage the previous night, as well as heart rate variability at rest in the morning.

3. The data analysis-based dynamic planning method for rehabilitation pathways according to claim 1, characterized in that, The target data is processed to extract central and peripheral state features, including the following steps: When extracting central state features, the frame difference method and optical flow method are used for eye movement video streams to calculate the variance of saccadic velocity and eyelid closure degree; fundamental frequency perturbation is extracted from speech signals; and real-time heart rate is extracted from photoplethysmography pulse waves. When extracting peripheral state features, time-frequency transformation is performed on the surface electromyography signal to calculate the median frequency of the power spectral density, and the slope is obtained by linear fitting of the sequence to characterize the spectral redshift trend; and jerk is calculated on the inertial sensor data to characterize the smoothness of the movement.

4. The data analysis-based dynamic planning method for rehabilitation pathways according to claim 1, characterized in that, The calculation of the central fatigue index based on the aforementioned central state characteristics includes the following steps: A pre-trained logistic regression model is used, with peripheral state features as the input vector; The input vector is fed into the Sigmoid function to calculate the probability that the patient is in a state of mental fatigue; Multiplying the probability by 100 maps it to a scalar of 0-100, and the scalar is used as the central fatigue index.

5. The data analysis-based dynamic planning method for rehabilitation pathways according to claim 1, characterized in that, The peripheral fatigue index satisfies: ； In the formula, I p This indicates the peripheral fatigue index; α and β The weighting coefficients are calculated based on information entropy to reflect the differences in sensitivity of different patients to changes in electromyography or movement; Norm indicates normalization. Indicates the median frequency slope; Indicates the smoothness of motion.

6. The data analysis-based dynamic planning method for rehabilitation pathways according to claim 1, characterized in that, Setting a preset peripheral fatigue threshold for patients and dynamically adjusting the central fatigue threshold based on the baseline data includes the following steps: Peripheral fatigue values ​​are set based on the patient's limb function rating; Based on baseline data, the previous night's sleep quality score was selected and normalized to 0. In interval 1, the deviation of the ratio of low-frequency to high-frequency morning heart rate variability was selected and normalized to 0. In the interval 1, the normalized sleep quality score and deviation were used as correction factors; The correction factor is substituted into the linear bias formula to dynamically adjust the central fatigue threshold.

7. The data analysis-based dynamic planning method for rehabilitation pathways according to claim 6, characterized in that, The linear bias formula is as follows: Th c_current =Th c_default -[k1×(1-S sleep )+k2×D hrv ]; In the formula, Th c_current This represents the dynamically adjusted central fatigue threshold; Th c_default This represents the default alarm threshold for central nervous system fatigue; S sleep Indicates the previous night's sleep quality score; D hrv This indicates the deviation of the low-frequency to high-frequency ratio of morning heart rate variability; k1 and k2 are preset sensitivity coefficients.

8. The data analysis-based dynamic planning method for rehabilitation pathways according to claim 1, characterized in that, The real-time central fatigue index is compared with the central fatigue threshold, and the real-time peripheral fatigue index is compared with the peripheral fatigue threshold. Control commands are then output, including the following steps: If the central fatigue index is less than or equal to the central fatigue threshold and the peripheral fatigue index is less than or equal to the peripheral fatigue threshold, it is determined to be a suitable training state, and the current rehabilitation schedule is maintained. If the central fatigue index is greater than the central fatigue threshold and the peripheral fatigue index is less than or equal to the peripheral fatigue threshold, it is determined that central fatigue is dominant, and the first control command is output to trigger the circuit breaker mechanism for cognitive tasks. If the central fatigue index is less than or equal to the central fatigue threshold and the peripheral fatigue index is greater than the peripheral fatigue threshold, it is determined that peripheral fatigue is dominant, and the second control command is output to trigger the pause or resistance unloading of the resistance training module. If the central fatigue index is greater than the central fatigue threshold and the peripheral fatigue index is greater than the peripheral fatigue threshold, it is determined to be high-level combined fatigue, and all active training tasks are forcibly suspended.

9. The data analysis-based dynamic planning method for rehabilitation pathways according to claim 5, characterized in that, A multimodal time-series prediction model is constructed based on target data and decoupling analysis results to predict the remaining tolerance time of patients under the influence of central or peripheral fatigue thresholds, including the following steps: The model input layer is constructed by: building a dynamic input layer using central fatigue features and peripheral fatigue features to obtain a high-dimensional time series tensor; and building a static stream input layer using baseline data to obtain the daily baseline vector. Design a baseline-gated LSTM unit, including: mapping the baseline vector of the static stream input layer to a weight factor through a fully connected layer, and applying the weight factor to the forget gate activation function of the LSTM unit; The model output layer is constructed and connected to a regression analyzer. Based on the processing results of the baseline-gated LSTM unit, the remaining tolerance time is output in a single numerical form. The remaining tolerance time is used to predict the time it will take for the patient to reach the set peripheral fatigue threshold or the dynamically adjusted central fatigue threshold under the current training load.

10. The data analysis-based dynamic planning method for rehabilitation pathways according to claim 8, characterized in that, Based on control instructions and remaining tolerance time, dynamic scheduling adjustment instructions are generated to complete the dynamic planning of the patient's rehabilitation pathway, including the following steps: If central fatigue is determined based on the first control command, tasks with cognitive loads higher than the set cognitive load are suspended, and tasks with cognitive loads lower than the set cognitive load are executed. If peripheral fatigue is determined based on the second control command, resistance training of the patient's limbs is suspended and replaced with speech therapy, cognitive training or training of the unaffected limbs. Based on the remaining endurance time, the fatigue threshold is predicted, and based on the fatigue threshold, continuous training sessions are broken down into intermittent training sessions with rest periods inserted.

11. A data analysis-based dynamic programming method system for rehabilitation pathways, characterized in that, include: The acquisition unit is used to acquire the patient's target data, which includes baseline data, central nervous system status data, and peripheral nervous system status data. The analysis unit, used to perform decoupled analysis of patient fatigue characteristics based on target data, includes: The normalization module is used to process the target data and extract central state features and peripheral state features; The calculation module is used to calculate the central fatigue index based on the central state characteristics and the peripheral fatigue index based on the peripheral state characteristics. An adjustment module is used to preset the patient's peripheral fatigue threshold and dynamically adjust the patient's central fatigue threshold based on the baseline data. The output module is used to compare the real-time central fatigue index with the central fatigue threshold and the real-time peripheral fatigue index with the peripheral fatigue threshold, and output control commands. The prediction unit is used to construct a multimodal time-series prediction model based on target data and decoupling analysis results to predict the patient's remaining tolerance time under the influence of central or peripheral fatigue thresholds. The planning unit is used to generate dynamic scheduling adjustment instructions based on control commands and remaining tolerance time, thereby completing the dynamic planning of the patient's rehabilitation path.

12. The data analysis-based dynamic planning method system for rehabilitation pathways according to claim 11, characterized in that, The prediction unit includes: The first input construction module is used to construct the model input layer, including: constructing a dynamic input layer through central fatigue features and peripheral fatigue features to obtain a high-dimensional time series tensor; and constructing a static stream input layer through baseline data to obtain the daily baseline vector. The second feature fusion and extraction module is used to design the baseline-gated LSTM unit, including: mapping the baseline vector of the static flow input layer to a weight factor through a fully connected layer, and applying the weight factor to the forget gate activation function of the LSTM unit to output a temporal feature hidden vector that incorporates individual baseline weights; The third time prediction output module is used to construct the model output layer. Its input is the time-series feature hidden vector output by the second module. By connecting to the regression analyzer, based on the processing results of the baseline-gated LSTM unit, it outputs the remaining tolerance time in a single numerical form. The remaining tolerance time is used to predict the time when the patient will reach the set peripheral fatigue threshold or the dynamically adjusted central fatigue threshold under the current training load.

13. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, when executing a computer program stored in memory, implements the data analysis-based dynamic programming method for rehabilitation pathways as described in any one of claims 1-10.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the data analysis-based dynamic planning method for rehabilitation pathways as described in any one of claims 1-10.

Images