Non-electric variable intelligent feedback control system of electric standing bed for severe stroke rehabilitation

Abstract

The application relates to the technical field of medical instrument control, in particular to a non-electric variable intelligent feedback regulation system of a stroke severe rehabilitation electric standing bed, wherein the system comprises a signal acquisition module which acquires physiological variable data through a non-invasive sensing array; a signal analysis module which extracts a multi-dimensional feature vector, carries out recursive analysis on the multi-dimensional feature vector based on a time sequence prediction model, generates a physiological state trend curve, and analyzes a current physiological reserve threshold and an expected imbalance probability of a patient according to a preset stroke severe body position tolerance model; a standing regulation module which is used for adaptively regulating the standing angle and angular velocity of the electric standing bed; and a safety protection and emergency disposal module which is used for generating a multi-level hierarchical early warning signal when the physiological state trend curve reaches a preset risk threshold or triggers safety protection, and generating a corresponding hierarchical protection strategy according to the multi-level hierarchical early warning signal. Thus, the problems of lagging control mode, high false alarm rate and passive risk response are solved.

Description

Technical Field

[0001] This application relates to the field of medical device control technology, and in particular to a non-electric variable intelligent feedback control system for an electric standing bed for severe stroke rehabilitation. Background Technology

[0002] Severe stroke patients often experience altered consciousness, hemiplegia, abnormal muscle tone, and unstable respiratory and circulatory functions. Prolonged bed rest can easily lead to various complications such as pressure sores, deep vein thrombosis in the lower extremities, pulmonary infections, and joint contractures, severely impacting patient prognosis. Electric standing beds, as key equipment for early rehabilitation of critically ill patients, can help patients achieve progressive upright training through positional changes, which is of significant clinical importance in improving cardiopulmonary function and promoting neurological recovery.

[0003] Existing electric standing beds mostly employ open-loop control with fixed angles and speeds or simple timed programs, relying on manual adjustments and experience-based judgment by medical staff. This makes it difficult to adaptively adjust to the patient's real-time physiological state, lacking awareness and feedback on the patient's physiological condition. While some devices integrate basic vital sign monitoring functions such as heart rate and blood pressure, they often use threshold-triggered alarms (e.g., audible and visual alarms when blood pressure is below 90 / 60 mmHg), which is a reactive, post-event response. They issue warnings only when a risk has occurred or is about to occur, failing to proactively intervene in the early stages of a risk. This approach has several shortcomings: First, threshold alarms based on single indicators are susceptible to individual differences and instantaneous fluctuations, resulting in a high false alarm rate and "alarm fatigue" among medical staff. Second, the lack of fusion analysis of multimodal physiological signals makes it impossible to capture early trend characteristics of risks such as postural tolerance and spasticity. Third, alarm information remains merely at the notification level, failing to integrate with the bed's motion control system, thus hindering automatic emergency response when risks occur and still requiring manual operation, posing a risk of response delay. Summary of the Invention

[0004] This application provides a non-electric variable intelligent feedback control system for an electric standing bed for severe stroke rehabilitation, in order to solve problems such as lagging control methods, high false alarm rates, and passive risk response.

[0005] The first aspect of this application provides a non-electric variable intelligent feedback control system for an electric standing bed in the rehabilitation of stroke patients, comprising: a signal sensing module, a signal analysis module, a standing control module, and a safety protection and emergency response module. The signal sensing module is used to collect physiological variable data through a non-invasive sensor array, including patient body surface pressure distribution, respiratory rhythm, and muscle tone. The signal analysis module is used to extract features from the physiological variable data to obtain multi-dimensional feature vectors, and to perform recursive analysis on the multi-dimensional feature vectors based on a time-series prediction model to generate a physiological state trend curve. The system analyzes the patient's current physiological reserve threshold and expected imbalance probability based on a preset postural tolerance model for severe stroke. The standing control module adaptively adjusts the standing angle and angular velocity of the electric standing bed according to the physiological reserve threshold and expected imbalance probability. When the expected imbalance probability exceeds the safe range or the physiological reserve threshold is insufficient, a safety protection is triggered. The safety protection and emergency response module generates a multi-level hierarchical early warning signal when the physiological state trend curve reaches a preset risk threshold or when the standing control module triggers the safety protection. Based on the multi-level hierarchical early warning signal, a corresponding graded protection strategy is generated.

[0006] Optionally, the signal sensing module includes: a body surface pressure sensing unit, a respiratory rhythm sensing unit, a muscle tension sensing unit, and a data preprocessing unit. The body surface pressure sensing unit collects static pressure distribution images and dynamic pressure change sequences of the patient's contact with the bed surface using a distributed array of high-density flexible piezoresistive sensors integrated into the mattress's weight-bearing area. It then extracts the body's center of gravity offset and body weight-bearing symmetry index in real time using a pressure center trajectory calculation algorithm. The respiratory rhythm sensing unit captures mechanical vibration signals caused by respiratory movements in the patient's chest and abdomen using a vibration sensing array embedded in the corresponding chest and abdominal areas of the mattress. After removing interference using an adaptive filtering algorithm, it extracts the respiratory rate, inspiratory / expiratory time ratio, and relative tidal volume change trends. The muscle tension sensing unit collects electromyographic signals during muscle activity using elastic strain gauge-type non-electrical sensing components attached to the patient's elbows and major muscle groups in the lower limbs. It extracts the root mean square value, median frequency, and electromyographic integral value through time-frequency domain analysis. The data preprocessing unit performs spatiotemporal alignment and missing value imputation on the data collected by the sensing units and then integrates the data to obtain physiological variable data.

[0007] Optionally, the signal parsing module includes: a feature extraction unit, a physiological state trend curve generation unit, and a postural tolerance estimation unit. The feature extraction unit receives physiological variable data, performs clinical feature mining and nonlinear dynamic feature extraction on it, and obtains a multidimensional feature vector. The physiological state trend curve generation unit builds a time-series prediction model based on an LSTM-RNN network, inputs the multidimensional feature vector into the trained time-series prediction model, outputs a sequence of predicted values ​​for each physiological indicator within a future preset time window, and generates a physiological state trend curve based on the predicted value sequence. The postural tolerance estimation unit builds a postural tolerance model for severe stroke, inputs the multidimensional feature vector and the physiological state trend curve into the trained postural tolerance model for severe stroke, performs personalized matching and association reasoning, and outputs the patient's current physiological reserve threshold and expected imbalance probability.

[0008] Optionally, the standing control module includes: an adaptive standing planning unit and a safe collaborative execution unit. The adaptive standing planning unit receives the physiological reserve threshold and the expected imbalance probability in real time, and dynamically obtains the optimal standing movement parameters through a multi-objective optimization algorithm based on a preset rehabilitation training target curve. The standing movement parameters include the target angle, angular velocity curve, and pause point. The safe collaborative execution unit converts the standing movement parameters into motor control commands to drive the electric standing bed, and simultaneously monitors the physiological reserve threshold and the expected imbalance probability. When the expected imbalance probability exceeds the safe range or the physiological reserve threshold is insufficient, it switches to a safety protection mode. The safety protection mode restores the bed to a supine position at the fastest safe speed and keeps the position locked until manually released.

[0009] Optionally, the safety protection and emergency response module includes: a risk grading and early warning unit, a graded protection strategy generation unit, and an emergency response execution unit. The risk grading and early warning unit receives physiological state trend curves, physiological reserve thresholds, and expected imbalance probabilities. Based on a preset multi-dimensional risk quantification assessment matrix, it classifies risk events into three levels and generates corresponding early warning signals. The graded protection strategy generation unit receives the early warning signals and, through a built-in rule engine decision model, generates a graded protection strategy that matches the current risk level and the patient's individual characteristics. The emergency response execution unit converts the graded protection strategy into an executable instruction sequence and drives the bed structure and auxiliary devices to perform protective actions according to the instruction sequence.

[0010] Optionally, the construction and operation of the postural tolerance model for severe stroke includes the following steps: collecting and structuring multi-source medical knowledge in the field of severe stroke, constructing a knowledge graph for postural tolerance assessment, wherein the nodes of the knowledge graph include patient static attributes, stroke lesion features, rehabilitation stage, and postural response, and the edges of the knowledge graph represent the medical semantic relationships between nodes; constructing a feature learner using a deep neural network with a multi-head attention mechanism, mining deep feature representations related to postural tolerance from the multi-dimensional feature vectors and the physiological state trend curves, and obtaining deep feature vectors; matching the knowledge graph with the current patient's individual characteristics and real-time rehabilitation stage, performing cross-modal alignment of the deep feature vectors with the matched knowledge graph entities, and aggregating the first-order and multi-hop neighbor information of entities through a graph attention network to generate a knowledge-enhanced patient representation; inputting the knowledge-enhanced patient representation into a multi-task learning output layer, and jointly optimizing the physiological reserve threshold regression task and the expected imbalance probability classification task by sharing hidden layers and task-specific branches, wherein the physiological reserve threshold regression task outputs the patient's current physiological reserve threshold, and the expected imbalance probability classification task outputs the patient's current expected imbalance probability.

[0011] The second aspect of this application provides a non-electric variable intelligent feedback control method for an electric standing bed for stroke rehabilitation, comprising the following steps: collecting physiological variable data through a non-invasive sensor array, the physiological variable data including patient body surface pressure distribution, respiratory rhythm, and muscle tension; extracting features from the physiological variable data to obtain multi-dimensional feature vectors, performing recursive analysis on the multi-dimensional feature vectors based on a time-series prediction model to generate a physiological state trend curve, and analyzing the patient's current physiological reserve threshold and expected imbalance probability according to a preset stroke rehabilitation posture tolerance model; adaptively controlling the standing angle and angular velocity of the electric standing bed according to the physiological reserve threshold and expected imbalance probability, triggering safety protection when the expected imbalance probability exceeds the safe range or the physiological reserve threshold is insufficient; generating multi-level hierarchical warning signals when the physiological state trend curve reaches a preset risk threshold or the standing control module triggers safety protection, and generating corresponding graded protection strategies based on the multi-level hierarchical warning signals.

[0012] A third aspect of this application provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform the non-electric variable intelligent feedback control method for an electric standing bed for stroke rehabilitation as described in the above embodiments.

[0013] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the non-electric variable intelligent feedback control method for an electric standing bed for severe stroke rehabilitation as described in the above embodiments.

[0014] The fifth aspect of this application provides a computer program product that stores a computer program that, when executed by a processor, implements the non-electric variable intelligent feedback control method for an electric standing bed for severe stroke rehabilitation as described in the above embodiments.

[0015] The beneficial effects of using the present invention are as follows:

[0016] This application embodiment achieves real-time, non-invasive, and stable perception of the postural tolerance status of critically ill stroke patients through non-electrophysiological variables such as body surface pressure distribution, respiratory rhythm, and muscle tension. After in-depth mining and intelligent prediction by the signal analysis module, the standing control module dynamically adjusts the standing angle and angular velocity to match the bed movement with the patient's physiological state in real time, significantly improving the personalization and safety of rehabilitation training. The signal analysis module performs recursive analysis of the physiological state trend curve through a time-series prediction model, and combines it with the postural tolerance model for critically ill stroke patients to analyze the patient's current physiological reserve threshold and expected imbalance probability. This allows for early identification and proactive intervention at the risk bud stage, gaining valuable golden time for clinical treatment. The standing control module adjusts the standing angle and angular velocity according to the patient's... The system dynamically adjusts the standing angle, angular velocity, and pause duration based on real-time tolerance levels, achieving a precise match between rehabilitation training intensity and the patient's physiological capacity. This ensures the effectiveness of rehabilitation training while avoiding safety risks caused by sudden changes in posture, muscle spasms, and shifts in the center of gravity. It is particularly suitable for critically ill patients with impaired consciousness or who cannot actively express discomfort. By constructing an integrated safety mechanism for risk-level early warning and tiered protection strategies, the system deeply links early warning signals with bedside movement control. Tiered protection actions can be automatically initiated at the initial stage of risk, and if necessary, the patient can quickly return to a supine position and lock the position without relying on manual intervention. This significantly shortens emergency response delays and maximizes the safety of critically ill stroke patients during postural training. This solves problems such as delayed control methods, high false alarm rates, and passive risk responses.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0019] Figure 1This is a schematic diagram of the non-electric variable intelligent feedback control system for an electric standing bed for severe stroke rehabilitation provided according to an embodiment of this application;

[0020] Figure 2 This is a flowchart of the non-electric variable intelligent feedback control method for an electric standing bed for severe stroke rehabilitation provided in the embodiments of this application;

[0021] Figure 3 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0022] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0023] The following description, with reference to the accompanying drawings, describes an embodiment of the intelligent feedback control system for non-electrical variables of an electric standing bed for stroke rehabilitation in severe cases. Addressing the issues of lagging control methods, high false alarm rates, and passive risk response mentioned in the background art, this application provides an intelligent feedback control system for non-electrical variables of an electric standing bed for stroke rehabilitation in severe cases. In this system, non-electrical physiological variables such as body surface pressure distribution, respiratory rhythm, and muscle tension are used to achieve real-time, non-invasive, and stable perception of the postural tolerance status of stroke patients. After in-depth analysis and intelligent prediction by the signal analysis module, the standing control module dynamically adjusts the standing angle and angular velocity, ensuring that the bed movement matches the patient's physiological state in real time, significantly improving the personalization and safety of rehabilitation training. The signal analysis module uses a time-series prediction model to recursively analyze the physiological state trend curve, and combines this with a stroke severe case postural tolerance model to determine the patient's current physiological reserve threshold and expected probability of imbalance, allowing for early intervention at the risk bud stage. Identifying and proactively intervening buys precious golden time for clinical treatment. The standing control module dynamically adjusts the standing angle, angular velocity, and pause duration based on the patient's real-time tolerance, achieving a precise match between rehabilitation training intensity and the patient's physiological capacity. This ensures the effectiveness of rehabilitation training while avoiding safety risks caused by sudden changes in posture, muscle spasms, and center of gravity shifts, making it particularly suitable for critically ill patients with impaired consciousness or who cannot actively express discomfort. By constructing an integrated safety mechanism for risk-level early warning and graded protection strategies for emergency response, the system deeply links early warning signals with bed movement control. It automatically initiates graded protection actions at the initial stage of risk, and if necessary, quickly returns the patient to a supine position and locks the position, eliminating the need for manual intervention. This significantly shortens emergency response delays and maximizes the safety of critically ill stroke patients during postural training. This solves problems such as delayed control methods, high false alarm rates, and passive risk response.

[0024] Specifically, Figure 1 This is a schematic diagram of the non-electric variable intelligent feedback control system for the electric standing bed for severe stroke rehabilitation provided in this embodiment of the application.

[0025] like Figure 1 As shown, the non-electric variable intelligent feedback control system 10 of the electric standing bed for severe stroke rehabilitation includes: a signal sensing module 100, a signal analysis module 200, a standing control module 300, and a safety protection and emergency response module 400.

[0026] The system includes a signal sensing module 100 for collecting physiological variable data via a non-invasive sensor array, including patient surface pressure distribution, respiratory rhythm, and muscle tension. A signal analysis module 200 extracts features from the physiological variable data to obtain multi-dimensional feature vectors, performs recursive analysis on these vectors based on a time-series prediction model to generate a physiological state trend curve, and analyzes the patient's current physiological reserve threshold and expected imbalance probability according to a preset postural tolerance model for severe stroke. A standing control module 300 adaptively adjusts the standing angle and angular velocity of the electric standing bed based on the physiological reserve threshold and expected imbalance probability. When the expected imbalance probability exceeds the safe range or the physiological reserve threshold is insufficient, a safety protection mechanism is triggered. A safety protection and emergency response module 400 generates multi-level hierarchical warning signals when the physiological state trend curve reaches a preset risk threshold or when the standing control module triggers safety protection. Based on these warning signals, a corresponding graded protection strategy is generated.

[0027] It is understood that the embodiments of this application achieve real-time, non-invasive, and stable perception of the postural tolerance status of critically ill stroke patients through non-electrophysiological variables such as body surface pressure distribution, respiratory rhythm, and muscle tension. After in-depth mining and intelligent prediction by the signal analysis module, the standing control module dynamically adjusts the standing angle and angular velocity to match the bed movement with the patient's physiological state in real time, significantly improving the personalization and safety of rehabilitation training. The signal analysis module performs recursive analysis of the physiological state trend curve through a time-series prediction model, and combines it with the postural tolerance model for critically ill stroke patients to analyze the patient's current physiological reserve threshold and expected imbalance probability. This allows for early identification and proactive intervention at the risk bud stage, gaining valuable golden time for clinical treatment. The standing control module is based on... The system dynamically adjusts the standing angle, angular velocity, and pause duration based on the patient's real-time tolerance, achieving a precise match between the intensity of rehabilitation training and the patient's physiological capacity. This ensures the effectiveness of rehabilitation training while avoiding safety risks caused by sudden changes in posture, muscle spasms, and shifts in the center of gravity. It is particularly suitable for critically ill patients with impaired consciousness or who are unable to actively express discomfort. By constructing an integrated safety mechanism that combines risk grading and early warning with tiered protection strategies and emergency response, the system deeply links early warning signals with bed movement control. Tiered protection actions can be automatically initiated at the initial stage of risk, and if necessary, the patient can be quickly reset to a supine position and locked in place without relying on manual intervention. This significantly shortens the emergency response delay and maximizes the safety of critically ill stroke patients during postural training.

[0028] In this embodiment, the signal sensing module includes: a body surface pressure sensing unit, a respiratory rhythm sensing unit, a muscle tension sensing unit, and a data preprocessing unit.

[0029] The system includes several components: a surface pressure sensing unit, a high-density flexible piezoresistive sensor array integrated into the mattress's weight-bearing area to collect static pressure distribution images and dynamic pressure change sequences between the patient and the bed surface, and an algorithm for calculating the pressure center trajectory to extract the body's center of gravity offset and body weight-bearing symmetry index in real time; a respiratory rhythm sensing unit, a vibration sensor array embedded in the corresponding chest and abdominal areas of the mattress to capture mechanical vibration signals caused by respiratory movements in the patient's chest and abdomen, and an adaptive filtering algorithm to remove interference and extract respiratory rate, inspiratory / expiratory time ratio, and relative tidal volume trends; a muscle tension sensing unit, elastic strain gauge-type non-electrical sensing components attached to the patient's elbows and major muscle groups in the lower limbs to collect electromyographic signals during muscle activity, and extracting root mean square values, median frequency, and electromyographic integral values ​​through time-frequency domain analysis; and a data preprocessing unit, which performs spatiotemporal alignment and missing value imputation on the data collected by the above sensing units before integrating the data to obtain physiological variable data.

[0030] Specifically, the surface pressure sensing unit employs a high-density flexible piezoresistive sensor array. This array is ergonomically zoned, deploying sensor nodes ranging from 16×16 to 32×32 in corresponding areas of the mattress: the shoulders and back, lumbosacral region, buttocks, and posterior lower limbs. The spacing between adjacent nodes is ≤2cm, ensuring the spatial resolution of pressure distribution acquisition meets the requirements for clinical pressure ulcer risk assessment. Each sensor node utilizes a nanocomposite piezoresistive material, exhibiting good linearity and sensitivity within a pressure range of 0-200mmHg. The surface pressure sensing unit continuously acquires the resistance values ​​of each node at a sampling frequency of at least 20Hz. After analog-to-digital conversion, a frame-format pressure distribution matrix is ​​generated, where each frame corresponds to a two-dimensional pressure distribution image at a given moment. Multiple consecutive frames constitute time-series data of pressure changes.

[0031] The specific implementation of the pressure center trajectory calculation algorithm is as follows:

[0032] For each frame of the pressure distribution image, first calculate the coordinates of the pressure center ( The calculation formula is:

[0033] in, The x-coordinate of the pressure center The ordinate of the pressure center is given by i and j, which are the row and column indices of the sensor array, respectively, m is the number of rows in the sensor array, and n is the number of columns in the sensor array. Let be the pressure value measured by the sensor located in the i-th row and j-th column. This represents the x-coordinate position of the sensor in the i-th row. This represents the ordinate position of the sensor in column j.

[0034] By using the coordinates of the pressure center over multiple consecutive frames, a trajectory curve of the pressure center changing over time is generated. Based on this, the Euclidean distance between the current pressure center and the geometric center of the bed surface is extracted as the body's center of gravity offset, used to quantify the degree to which the patient deviates from the center of the bed surface. The bed surface is divided into left and right halves, and the ratio of the total pressure on the left and right sides is calculated as a body weight-bearing symmetry index, used to quantify the weight-bearing difference between the healthy and affected sides of hemiplegic patients. An index close to 1 indicates symmetrical weight-bearing, while an index far from 1 indicates eccentric weight-bearing.

[0035] The vibration sensing array of the respiratory rhythm sensing unit employs miniature piezoelectric vibration sensors embedded in the mattress in the area corresponding to the patient's chest and abdomen. It does not require contact with the patient's skin and captures respiratory motion signals through mechanical vibrations transmitted from the mattress. This respiratory motion signal is a mixed signal containing respiratory components and various interference sources, including: low-frequency baseline drift caused by changes in bed angle during standing, high-frequency transient interference caused by patient limb movements or medical staff operations, and 50Hz power frequency interference. An adaptive filtering algorithm is used for signal processing: through a minimum mean square error adaptive filtering structure, baseline drift, transient impacts, and power frequency interference in the mixed signal are dynamically canceled as noise reference components in sequence, achieving high-precision reconstruction of the pure respiratory wave while preserving the characteristics of the respiratory waveform. From the reconstructed respiratory wave, a peak detection algorithm identifies the inspiratory initiation, inspiratory peak, expiratory initiation, and expiratory peak, and then calculates the respiratory rate, inspiratory / expiratory time ratio, and relative tidal volume change trend. The relative tidal volume change trend is characterized by the relative rate of change of respiratory wave amplitude, used to assess the consistency of respiratory depth.

[0036] The elastic strain gauge-type non-electric sensing component of the muscle tension sensing unit uses a highly sensitive elastic strain material that can be fitted and fixed to the surface of major muscle groups such as the elbow flexors / extensors of the upper limb, the quadriceps femoris and gastrocnemius of the lower limb. The tension generated during muscle contraction causes the strain gauge to undergo mechanical deformation, converting the change in muscle tension into a strain signal. Then, through time-frequency domain analysis, the root mean square value, median frequency, and electromyographic integral value are extracted. Among them, the root mean square value reflects the amplitude of muscle contraction, the median frequency reflects the degree of muscle fatigue, and the electromyographic integral value reflects the total energy of muscle contraction, realizing real-time monitoring of muscle spasms, muscle fatigue, and other states.

[0037] The data preprocessing unit employs a multi-source data synchronization and alignment algorithm. Based on the acquisition time sequence of the body surface pressure sensing unit, it performs spatiotemporal alignment on the data acquired by the respiratory rhythm sensing unit and the muscle tension sensing unit to ensure that the timestamps of data from different dimensions are consistent. To address potential signal loss during sensor acquisition, an interpolation algorithm based on adjacent time-time data is used to fill in missing values. Finally, the data acquired by each sensing unit is integrated into physiological variable data in a unified format and output to the signal analysis module for subsequent feature extraction and analysis.

[0038] In this embodiment, the signal parsing module includes: a feature extraction unit, a physiological state trend curve generation unit, and a body position tolerance estimation unit.

[0039] The system comprises the following components: a feature extraction unit receives physiological variable data, performs clinical feature mining and nonlinear dynamic feature extraction to obtain multidimensional feature vectors; a physiological state trend curve generation unit builds a time-series prediction model based on an LSTM-RNN network, inputs the multidimensional feature vectors into the trained time-series prediction model, outputs the predicted value sequence of each physiological indicator within a preset future time window, and generates a physiological state trend curve based on the predicted value sequence; and a postural tolerance estimation unit builds a postural tolerance model for severe stroke, inputs the multidimensional feature vectors and physiological state trend curves into the trained postural tolerance model for severe stroke for personalized matching and correlation reasoning, and outputs the patient's current physiological reserve threshold and expected imbalance probability.

[0040] Specifically, the feature extraction unit receives physiological variable data after spatiotemporal alignment and missing value imputation, and performs multi-dimensional clinical feature mining and nonlinear dynamic feature extraction on it, including:

[0041] From the time-series data of body position center of gravity offset and body weight-bearing symmetry index output by the body surface pressure sensing unit, the root mean square of the swing velocity of the pressure center swing trajectory, the area of ​​the 95% confidence ellipse, and the sample entropy are extracted and used to quantify the patient's agility in adjusting body position, static balance maintenance ability, and the regularity of autonomous regulation.

[0042] From the time-series data of respiratory rate, inspiratory / expiratory time ratio and relative tidal volume change trends output by the respiratory rhythm sensing unit, the instantaneous frequency variability coefficient, inspiratory / expiratory ratio drift rate and tidal volume fluctuation amplitude of the respiratory rhythm are extracted and used to evaluate the stability of respiratory center drive, the trend of airway resistance change and the degree of respiratory muscle fatigue, respectively.

[0043] From the root mean square value, median frequency, and electromyographic integral value time series data output by the muscle tension sensing unit, the wavelet packet decomposition of the electromyographic signal is used to extract the energy proportion of each frequency band, the instantaneous change rate of the median frequency, and the electromyographic burst interval time, which are respectively used to identify the frequency domain drift characteristics in the pre-spasticity phase, quantify the dynamic change rate of muscle tension, and determine the active muscle strength arousal mode.

[0044] By performing principal component analysis to reduce dimensionality and normalize the above-mentioned multidimensional clinical features, a multidimensional fusion feature vector that can comprehensively characterize the patient's current neuromuscular state and physiological compensatory ability is constructed.

[0045] The time-series prediction model built on an LSTM-RNN network consists of an input layer, an LSTM hidden layer, an attention mechanism layer, and an output layer. The input layer receives multi-dimensional fused feature vectors from T consecutive time windows. The LSTM hidden layer consists of 2-3 stacked LSTM layers, each containing 64-128 memory units. Each LSTM unit includes a forget gate, an input gate, and an output gate. The attention mechanism layer weights and fuses the hidden states at different time steps, enabling the model to automatically focus on the most critical historical moments for prediction. The output layer uses a fully connected network to output the predicted value sequence of key physiological indicators within a preset future time window. The output layer has dimensions (P, M), where P is the number of prediction steps and M is the number of predicted indicators (including key indicators such as center of pressure oscillation velocity, respiratory rate, and median electromyography frequency).

[0046] Supervised learning was employed for training, with training data derived from historically collected postural training data of severely ill stroke patients. To enhance the robustness of predictions, a rolling iterative prediction approach was used: in each control cycle, the model outputs a prediction sequence for the next P steps based on the latest T historical time windows of data; as new measured data arrives, the model continuously updates the prediction results.

[0047] Based on the predicted value sequence, the physiological state trend curve generation unit synthesizes multidimensional physiological state trend curves, including the pressure center trajectory prediction curve, respiratory rhythm trend curve, and electromyographic characteristic trend curve.

[0048] In this embodiment of the application, the construction and operation of the postural tolerance model for severe stroke includes the following steps:

[0049] Multi-source medical knowledge in the field of severe stroke is collected and structured to construct a knowledge graph for postural tolerance assessment. The nodes of the knowledge graph include patient static attributes, stroke lesion characteristics, rehabilitation stage, and postural response. The edges of the knowledge graph represent the medical semantic relationships between nodes.

[0050] A feature learner is built using a deep neural network with a multi-head attention mechanism. Deep feature representations related to postural tolerance are extracted from multi-dimensional feature vectors and physiological state trend curves to obtain deep feature vectors.

[0051] Based on the current individual characteristics of patients and the knowledge graph matching the real-time rehabilitation stage, the deep feature vectors are aligned with the entities in the matching knowledge graph across modalities. The first-order and multi-hop neighbor information of the entities is aggregated through a graph attention network to generate knowledge-enhanced patient representations.

[0052] The knowledge-enhanced patient representation is input into the multi-task learning output layer. By sharing the hidden layer and task-specific branches, the physiological reserve threshold regression task and the expected imbalance probability classification task are jointly optimized. The physiological reserve threshold regression task outputs the patient's current physiological reserve threshold, and the expected imbalance probability classification task outputs the patient's current expected imbalance probability.

[0053] Specifically, multi-source knowledge in the field of severe stroke is collected and integrated from structured clinical guidelines, electronic medical records, medical literature, and expert systems to construct a knowledge graph for postural tolerance assessment. The entity types of the knowledge graph nodes include patient static attributes, stroke lesion characteristics, rehabilitation stage, and postural responses. Patient static attributes include age, gender, and underlying diseases; stroke lesion characteristics include type, location, and NIHSS score; and postural responses include susceptibility to orthostatic hypotension, historical extreme values ​​of standing tolerance angles, and spasticity trigger thresholds. Edges in the knowledge graph represent medical semantic relationships between nodes. The graph is stored in graph databases such as Neo4j, supporting efficient graph traversal and relational reasoning.

[0054] A multi-head attention deep neural network based on the Transformer encoder architecture is used as the feature learner. The physiological state trend curve is temporally encoded through one-dimensional convolution and positional encoding to obtain a fixed-length physiological state trend curve encoding vector. This encoding vector is then concatenated and fused with a multi-dimensional feature vector and input into the feature learner. The feature weights are calculated through a multi-head attention mechanism to extract core features highly correlated with postural tolerance. After processing through a multi-layer fully connected network and a non-linear activation function, a deep feature vector is obtained.

[0055] Based on the patient's current individual characteristics and real-time rehabilitation stage, matching is performed in the knowledge graph to obtain matched knowledge graph entities. The deep feature vectors and the matched knowledge graph entities are then aligned across modalities using a gating mechanism. The gating fusion formula is as follows:

[0056] in, is the cross-modal aligned feature vector; g is the gating coefficient; It is a depth feature vector; This is the mapping matrix for entity features in the knowledge graph; For matching knowledge graph entities; The weight matrix for a gated fully connected network; For bias terms; This is the Sigmoid activation function.

[0057] The graph attention network aggregates the first-order and multi-hop neighbor information of entities, and the graph attention update formula is as follows: in, Let i be the set of neighbors of entity i; Let i be the node feature of entity i; Let be the node features of entity j; It is the attention coefficient; It is the characteristic transformation matrix; This is the Sigmoid activation function.

[0058] Through multi-layer stacking, information propagates in the graph, and finally the subgraph representation that aggregates relevant medical knowledge is fused with deep feature vectors to generate a knowledge-enhanced patient representation.

[0059] The knowledge-enhanced patient representation is input into a multi-task learning output layer. This multi-task learning output layer consists of a shared hidden layer and two task-specific branches: a physiological reserve threshold regression branch and an expected imbalance probability classification branch. The shared hidden layer is used for further feature extraction and optimization of the knowledge-enhanced patient representation, enhancing its feature representation capabilities. The expected imbalance probability classification branch uses a fully connected layer and a Softmax classifier, taking the shared hidden layer output as input and outputting the patient's current expected imbalance probability. The physiological reserve threshold regression branch uses a fully connected layer and linear activation, taking the shared hidden layer output as input and outputting a continuous value representing the patient's current physiological reserve threshold.

[0060] The two tasks are optimized simultaneously using a joint loss function. The formula for the joint loss function is as follows: in, For the joint loss function; The loss function for the physiological reserve threshold regression task; The loss function for the classification task with the expected imbalance probability; This is the loss weighting coefficient, with a value ranging from 0.4 to 0.6; This represents the number of training samples; Let be the predicted physiological reserve threshold for the r-th sample; Let be the clinical real physiological reserve threshold for the r-th sample; The true risk level label for the r-th sample (using one-hot encoding, c=1, 2, 3 correspond to low risk, medium risk, and high risk, respectively); The probability of predicting the r-th sample as having the c-th risk level.

[0061] In this embodiment, the standing control module includes: an adaptive standing planning unit and a safe collaborative execution unit.

[0062] The adaptive standing planning unit receives physiological reserve thresholds and expected imbalance probabilities in real time. Based on the preset rehabilitation training target curve, it dynamically obtains the optimal standing movement parameters through a multi-objective optimization algorithm. The standing movement parameters include target angle, angular velocity curve, and pause point. The safety collaborative execution unit converts the standing movement parameters into motor control commands to drive the electric standing bed. It simultaneously monitors physiological reserve thresholds and expected imbalance probabilities. When the expected imbalance probability exceeds the safe range or the physiological reserve threshold is insufficient, it switches to the safety protection mode. The safety protection mode restores the bed to a supine position at the fastest safe speed and keeps the position locked until manually released.

[0063] Specifically, the preset rehabilitation training target curve defines the expected angle-time relationship under an ideal risk-free state, which is represented by a piecewise increasing saturation curve.

[0064] The multi-objective optimization algorithm employs an improved NSGA-II algorithm, aiming to maximize rehabilitation stimulation and minimize physiological risk.

[0065] The probability minimization is the dual optimization objective, constrained by a physiological reserve threshold boundary, a motor operating safety threshold, and a patient posture comfort threshold. The optimal standing motion parameters are dynamically output through iterative optimization. The dual-objective optimization formula is as follows: in, Let u be a bi-objective optimization function, and u be a control sequence. The objective function for rehabilitation stimulation measures the deviation between the actual standing trajectory and the target curve of rehabilitation training. Its weight; Let the physiological risk objective function measure the physiological risk posed by the planned trajectory. Its weight; The current moment; To control the cycle; K is the number of prediction steps, and k is the time step index; The standing angle needs to be optimized; The angle corresponding to the rehabilitation training target curve; The square of the Euclidean distance; the future time. The predicted value of the expected imbalance probability at that moment. Its weight; This is the minimum permissible physiological reserve threshold; Predicted values ​​for physiological reserve thresholds at future moments; As a penalty for insufficient reserves, Its weight.

[0066] The system uses a 100Hz sampling frequency to synchronously collect real-time data on the changes in physiological reserve threshold and expected imbalance probability, and dynamically compares this data with a preset safety range. When the expected imbalance probability exceeds the preset safety range or the physiological reserve threshold falls below the preset safety threshold, the safety protection mode is immediately triggered. Without manual intervention, the system uses a motor reversal command to quickly and safely restore the bed to a 0° supine position. Simultaneously, a mechanical locking mechanism is activated to maintain the position locked, preventing accidental bed shaking, until medical staff manually release the lock after confirming the patient's safety.

[0067] Understandably, in this embodiment, the adaptive standing planning unit enables the bed to understand the patient's physiological tolerance and dynamically adjust the training intensity; the safe collaborative execution unit enables the bed to perceive risks and protect the patient in a timely manner. This design not only improves the safety and effectiveness of rehabilitation training but also provides a new technological paradigm for human-machine collaboration in the field of critical care rehabilitation.

[0068] In this embodiment, the security protection and emergency response module includes: a risk classification and early warning unit, a classification protection strategy generation unit, and an emergency response execution unit.

[0069] The risk grading and early warning unit receives physiological state trend curves, physiological reserve thresholds, and expected imbalance probabilities. Based on a preset multi-dimensional risk quantification assessment matrix, it divides risk events into three levels and generates early warning signals corresponding to each level. The graded protection strategy generation unit receives the early warning signals and generates graded protection strategies that match the current risk level and individual patient characteristics through a built-in rule engine decision model. The emergency response execution unit converts the graded protection strategy into an executable instruction sequence and drives the bed structure and auxiliary devices to perform protective actions according to the instruction sequence.

[0070] Specifically, a multi-dimensional risk quantification assessment matrix is ​​constructed. This matrix uses deviation from the physiological reserve threshold, expected imbalance probability, and rate of change of the physiological state trend curve as assessment dimensions. Each dimension is divided into 5 quantification levels (levels 1-5). A comprehensive risk score is calculated by weighted summation, with the score range from 0 to 100. Based on the score, risk events are classified into three levels: low-risk, medium-risk, and high-risk. A score of 0-30 is classified as low-risk; a score of 31-60 as medium-risk; and a score of 61-100 as high-risk. Specifically, the low-risk level corresponds to an expected imbalance probability <5%, a physiological reserve threshold deviation <10%, and a stable trend curve without abnormalities; the medium-risk level corresponds to an expected imbalance probability of 5%-15%, a physiological reserve threshold deviation of 10%-20%, and slight abnormal fluctuations in the trend curve; and the high-risk level corresponds to an expected imbalance probability >15%, a physiological reserve threshold deviation ≥20%, and a deteriorating trend curve.

[0071] The built-in rule engine decision model pre-stores clinical rules, risk management guidelines, and patient individual characteristic association rules for postural training in severe stroke rehabilitation. After receiving the warning signal output by the risk grading and early warning unit, it first matches the association rules between the current patient's individual characteristics and risk level, and then generates a graded protection strategy adapted to the current risk level and the patient's individual characteristics through the forward reasoning of the rule engine. The patient's individual characteristics include age, stroke lesion type, NIHSS score, and rehabilitation stage.

[0072] For example, a 68-year-old stroke patient with basal ganglia infarction and an NIHSS score of 8 is in the subacute phase of rehabilitation. During standing rehabilitation training, the risk grading and early warning unit receives the patient's physiological state trend curve, physiological reserve threshold, and expected imbalance probability in real time. Calculated using a multi-dimensional risk quantification assessment matrix, the physiological reserve threshold deviation is 15%, corresponding to quantification level 3 with a score of 35; the expected imbalance probability is 10%, also corresponding to quantification level 3 with a score of 35; and the rate of change of the physiological state trend curve is 0.8 units / second, corresponding to quantification level 2 with a score of 20. Therefore, the comprehensive risk score S = 0.4 × 35 + 0.35 × 35 + 0.25 × 20 = 31.25, classifying the patient as medium-risk. A flashing yellow indicator light and a low-volume beeping medium-risk warning signal are then generated. Upon receiving the warning signal, the graded protection strategy generation unit extracts the patient's individual characteristics, matches them with the association rules corresponding to the intermediate-risk level, and generates a protection strategy through forward reasoning: reduce the standing angular velocity from 1° / s to 0.2° / s, maintain the current 40° standing angle, and fine-tune the tightness of the lower limb straps (appropriately reducing strap pressure due to the patient's NIHSS score of 8). The risk score is updated every 5 seconds. The emergency response execution unit converts this strategy into motor control commands and strap adjustment commands, driving the bed to complete the angular velocity adjustment and angle maintenance, simultaneously adjusting the strap tightness, and collecting real-time data on the patient's muscle tension and respiratory rhythm, which is then fed back to the strategy generation unit. After 5 seconds, the overall risk score drops to 28 points (low-risk level).

[0073] It is understood that the embodiments of this application solve the technical pain points of traditional early warning systems, such as crudeness, false alarms and missed alarms, by constructing a quantitative and accurate multi-dimensional risk assessment system, and realize the scientific classification of risk levels and differentiated early warning; through the built-in rule engine decision model, personalized protection strategies are generated in combination with individual patient characteristics, breaking through the limitation of poor adaptability of traditional fixed strategies; through the coordinated linkage of the three units, an integrated closed-loop mechanism is constructed from risk quantitative assessment to personalized strategy generation and then to rapid emergency execution, which can complete the entire process of risk handling without manual intervention, and significantly reduce the workload of medical staff.

[0074] The following is a detailed description of the non-electric variable intelligent feedback control system for an electric standing bed in the rehabilitation of severe stroke patients, using a specific embodiment:

[0075] A 68-year-old patient with severe basal ganglia infarction and NIHSS score of 8, who was in the subacute rehabilitation phase and had unilateral hemiplegia, used the non-electric variable intelligent feedback control system of the stroke rehabilitation electric standing bed proposed in this application for standing rehabilitation training.

[0076] The signal sensing module collects patient physiological variable data through various sensing units: the body surface pressure sensing unit uses a high-density flexible piezoresistive sensor array in the mattress's weight-bearing area to collect static pressure distribution images and dynamic pressure change time sequences between the patient and the bed surface at a sampling frequency of 20Hz. Through a pressure center trajectory calculation algorithm, it extracts real-time data showing a body position center of gravity offset of 8mm and a body weight-bearing symmetry index of 0.65 (indicating insufficient weight-bearing on the affected side); the respiratory rhythm sensing unit uses miniature piezoelectric vibration sensors embedded in the chest and abdominal areas of the mattress to capture mechanical vibration signals generated by the patient's respiratory movements, which are then processed by an adaptive filtering algorithm. After removing low-frequency interference from bed angle changes, high-frequency transient interference from medical and nursing operations, and 50Hz power frequency interference, the respiratory rate was extracted as 18 breaths / minute, the inspiratory / expiratory time ratio was 1:1.8, and the tidal volume showed a relatively stable trend. The muscle tension sensing unit, through elastic strain gauge-type non-electrical sensing components attached to the elbow flexors of the upper limb and the quadriceps femoris and gastrocnemius muscles of the lower limb, collected electromyographic signals during muscle activity. Time-frequency domain analysis extracted a root mean square value of 0.3mV, a median frequency of 35Hz, and an integral electromyographic value of 1.2mV·s, indicating no obvious muscle spasm or fatigue. The data preprocessing unit, based on the acquisition sequence of the body surface pressure sensing unit, performed spatiotemporal alignment on the above three types of data, filled in one signal loss caused by slight patient positional shift, and integrated the standardized physiological variable data to the signal analysis module.

[0077] After receiving physiological variable data, the signal parsing module performs multi-dimensional feature mining and fusion. From the surface pressure-related data, it extracts the root mean square velocity of the pressure center's oscillation trajectory, the area of ​​the 95% confidence ellipse, and the sample entropy. From the respiratory rhythm data, it extracts the respiratory rate variability coefficient, the inspiratory-to-expiratory ratio drift rate, and the tidal volume fluctuation amplitude. From the muscle tension data, it extracts the energy proportion of each frequency band in the wavelet packet decomposition of electromyography (EMG) signals, the instantaneous rate of change of the median frequency, and the EMG burst interval. After principal component analysis for dimensionality reduction and normalization fusion, a multi-dimensional fused feature vector is generated. The physiological state trend curve generation unit inputs this multi-dimensional feature vector into a trained LSTM-RNN time series prediction model, outputting a predicted value sequence of each core physiological indicator for the next 30 seconds. Based on this, a pressure curve is synthesized. The force center trajectory prediction curve, respiratory rhythm trend curve, and electromyographic characteristic trend curve are used. The postural tolerance estimation unit inputs the multidimensional feature vector and physiological state trend curve into the postural tolerance model for severe stroke. The model first matches the patient's individual characteristics through a knowledge graph. The patient's individual characteristics are: 68 years old, basal ganglia infarction, NIHSS score of 8, and subacute phase. The deep feature vector and the matched knowledge graph entities are aligned across modalities through a gating mechanism. The graph attention network aggregates the entity neighbor information to generate a knowledge-enhanced patient representation. Then, through multi-task learning, the output layer is jointly optimized to finally output the patient's current physiological reserve threshold (pressure tolerance threshold 45°, respiratory tolerance threshold 20 breaths / minute, muscle tension safety threshold 0.5mV) and the expected imbalance probability (10%).

[0078] After receiving the physiological reserve threshold and the expected probability of imbalance, the adaptive standing planning unit performs multi-objective optimization using the improved NSGA-II algorithm based on the preset segmented incremental saturation rehabilitation training target curve. The dual objectives are maximizing rehabilitation stimulation and minimizing the probability of physiological risk. The constraints are the physiological reserve threshold boundary, the motor operation safety threshold, and the patient's body position comfort threshold. The optimal standing motion parameters are dynamically output: target angle 40°, angular velocity curve initial stage 0.5° / s, decreasing to 0.3° / s when approaching the target angle, and pause points set at 15° and 30°, with each pause lasting 60 seconds. The safety collaborative execution unit converts the above motion parameters into motor control commands to drive the electric standing bed to move smoothly, while simultaneously monitoring the patient's physiological reserve threshold and the expected probability of imbalance at a sampling frequency of 100Hz.

[0079] When the patient sits up to 40° and maintains this position for 20 seconds, the risk grading and early warning unit of the safety protection and emergency response module receives the physiological state trend curve, physiological reserve threshold, and expected imbalance probability in real time. Based on the multi-dimensional risk quantification assessment matrix, it calculates a comprehensive risk score: physiological reserve threshold deviation of 15% (corresponding to quantification level 3, score 35 points), expected imbalance probability of 10% (corresponding to quantification level 3, score 35 points), and physiological state trend curve change rate of 0.8 units / second (corresponding to quantification level 2, score 20 points). The comprehensive risk score S = 0.4×35 + 0.35×35 + 0.25×20 = 14 + 12.25 + 5 = 31.25 points, which is determined to be a medium-risk level. A medium-risk early warning signal is then generated, with a flashing yellow indicator light and a low-volume beeping sound. After receiving the early warning signal, the graded protection strategy generation unit uses its built-in rule engine decision model to match the patient's individual characteristics with the association rules of the intermediate-risk level, generating a graded protection strategy: reducing the standing angular velocity from 0.3° / s to 0.2° / s, maintaining the current 40° standing angle, fine-tuning the tightness of the lower limb straps, and updating the risk score every 5 seconds. The emergency response execution unit translates this strategy into motor control commands and strap adjustment commands, driving the bed to complete the angular velocity adjustment and angle maintenance, simultaneously adjusting the strap tightness, and collecting real-time data on the patient's muscle tension and respiratory rhythm, feeding it back to the strategy generation unit. After 5 seconds, the system recalculates the comprehensive risk score. As the patient's physiological state stabilizes, the score drops to 28 points, corresponding to a low-risk level.

[0080] In summary, the embodiments of this application achieve accurate and non-invasive acquisition of patient physiological data through fully non-invasive sensing, realize early prediction of patient physiological status and personalized quantification of postural tolerance by using LSTM-RNN time series prediction and postural tolerance model for severe stroke, realize dynamic adaptive control of standing movement parameters through improved NSGA-II multi-objective optimization algorithm, and realize accurate early warning and rapid treatment of risks by relying on multi-dimensional risk grading and personalized protection strategies. The entire process can balance the effectiveness of rehabilitation training and patient safety without human intervention, and significantly reduce the workload of medical staff.

[0081] Next, referring to the accompanying drawings, a non-electric variable intelligent feedback control method for an electric standing bed for severe stroke rehabilitation, according to an embodiment of this application, is described.

[0082] Specifically, Figure 2 The flowchart illustrates the non-electric variable intelligent feedback control method for the electric standing bed for severe stroke rehabilitation provided in this application embodiment.

[0083] like Figure 2 As shown, the non-electric variable intelligent feedback control method for the electric standing bed for severe stroke rehabilitation includes the following steps:

[0084] In step S101, physiological variable data are collected through a non-invasive sensor array. The physiological variable data includes the patient's body surface pressure distribution, respiratory rhythm, and muscle tension.

[0085] Specifically, a high-density flexible piezoresistive sensor array integrated into the mattress's weight-bearing area collects static pressure distribution images and dynamic pressure change time sequences between the patient and the bed surface. A pressure center trajectory calculation algorithm is used to extract the body's center of gravity offset and body weight-bearing symmetry index in real time. A vibration sensor array embedded in the corresponding chest and abdominal areas of the mattress captures mechanical vibration signals caused by respiratory movements in the patient's chest and abdomen. An adaptive filtering algorithm is used to remove interference before extracting respiratory rate, inspiratory / expiratory time ratio, and relative tidal volume trends. Elastic strain gauge-type non-electrical sensing components attached to the patient's elbows and major muscle groups in the lower limbs collect electromyographic signals during muscle activity. Time-frequency domain analysis is used to extract the root mean square value, median frequency, and electromyographic integral value. The collected data undergoes spatiotemporal alignment and missing value imputation before data integration to obtain physiological variable data.

[0086] It is understood that the embodiments of this application adopt a non-invasive sensing architecture, and through the collaborative acquisition of multiple types of sensing units, it realizes the comprehensive, real-time and accurate acquisition of three core physiological variables: patient body surface pressure, respiratory rhythm and muscle tension. At the same time, through a scientific data preprocessing process, it ensures the integrity, accuracy and consistency of physiological variable data, and lays a solid data foundation for the smooth implementation of the entire regulation method.

[0087] In step S102, feature extraction is performed on the physiological variable data to obtain multidimensional feature vectors. Based on the time series prediction model, the multidimensional feature vectors are recursively analyzed to generate physiological state trend curves. The patient's current physiological reserve threshold and expected imbalance probability are analyzed according to the preset postural tolerance model for severe stroke.

[0088] Specifically, a time-series prediction model is built based on an LSTM-RNN network. Multidimensional feature vectors are input into the trained model, which outputs a sequence of predicted values ​​for various physiological indicators within a preset future time window. A physiological state trend curve is generated based on this predicted value sequence. A postural tolerance model for severe stroke is also built. Multidimensional feature vectors and the physiological state trend curve are input into the trained model for personalized matching and correlation inference, outputting the patient's current physiological reserve threshold and the expected probability of imbalance.

[0089] It is understood that the embodiments of this application realize the early prediction of physiological state through the LSTM-RNN time series prediction model, and combined with the body position tolerance model that integrates medical knowledge, it realizes the accurate and personalized output of physiological reserve threshold and expected imbalance probability, which solves the pain point of the prior art that cannot accurately quantify the patient's body position tolerance ability.

[0090] In step S103, the standing angle and angular velocity of the electric standing bed are adaptively adjusted according to the physiological reserve threshold and the expected imbalance probability. When the expected imbalance probability exceeds the safe range or the physiological reserve threshold is insufficient, the safety protection is triggered.

[0091] Specifically, the system receives physiological reserve thresholds and expected imbalance probabilities in real time. Based on a preset rehabilitation training target curve, it dynamically obtains the optimal standing movement parameters through a multi-objective optimization algorithm. These parameters include the target angle, angular velocity curve, and pause point. The standing movement parameters are then converted into motor control commands to drive the electric standing bed. Simultaneously, the system monitors physiological reserve thresholds and expected imbalance probabilities. When the expected imbalance probability exceeds the safe range or the physiological reserve threshold is insufficient, it switches to a safety protection mode. This mode restores the bed to a supine position at the fastest safe speed and maintains the locked position until manually released.

[0092] It is understood that the embodiments of this application realize personalized, dynamic, and adaptive control of the standing movement parameters. It can dynamically adjust the standing angle, angular velocity, and pause duration according to the patient's real-time physiological reserve threshold and expected probability of imbalance, so as to ensure that the intensity of rehabilitation training is highly matched with the patient's physiological tolerance.

[0093] In step S104, when the physiological state trend curve reaches the preset risk threshold or the standing control module triggers safety protection, a multi-level hierarchical early warning signal is generated, and a corresponding hierarchical protection strategy is generated based on the multi-level hierarchical early warning signal.

[0094] Specifically, the system receives physiological state trend curves, physiological reserve thresholds, and expected imbalance probabilities. Based on a pre-set multi-dimensional risk quantification assessment matrix, it classifies risk events into three levels and generates corresponding warning signals. Through a built-in rule engine decision model, it generates a tiered protection strategy that matches the current risk level and the patient's individual characteristics. The tiered protection strategy is then translated into an executable sequence of instructions, which drives the bed structure and auxiliary devices to perform protective actions.

[0095] It is understood that the embodiments of this application construct an integrated safety response mechanism from risk warning to strategy generation to action execution, achieving precise risk level classification through multi-dimensional risk quantification assessment. Simultaneously, based on a rule engine decision model, it generates tiered protection strategies adapted to risk levels and individual patient characteristics, enabling personalized treatment for patients with different risk levels and individual characteristics. By rapidly converting protection strategies into executable instructions, the timeliness and accuracy of emergency response are ensured, further improving the safety assurance system of the entire control method and adapting to the clinical characteristics of severe stroke patients with significant individual differences and complex conditions.

[0096] The non-electric variable intelligent feedback control method for the electric standing bed for severe stroke rehabilitation proposed in this application realizes real-time, non-invasive, and stable perception of the postural tolerance status of severe stroke patients through non-electric physiological variables such as body surface pressure distribution, respiratory rhythm, and muscle tension. After in-depth mining and intelligent prediction by the signal analysis module, the standing control module dynamically adjusts the standing angle and angular velocity, so that the bed movement matches the patient's physiological state in real time, significantly improving the personalization and safety of rehabilitation training. The signal analysis module performs recursive analysis of the physiological state trend curve through a time-series prediction model, and combines it with the postural tolerance model for severe stroke to analyze the patient's current physiological reserve threshold and expected probability of imbalance. This allows for early identification and proactive intervention at the risk bud stage, gaining valuable time for clinical treatment. The "golden time" (or "golden window") is achieved through a dynamic adjustment module that dynamically adjusts the standing angle, angular velocity, and pause duration based on the patient's real-time tolerance. This ensures a precise match between the intensity of rehabilitation training and the patient's physiological capacity, guaranteeing the effectiveness of rehabilitation training while avoiding safety risks caused by sudden changes in posture, muscle spasms, or shifts in the center of gravity. This is particularly suitable for critically ill patients with impaired consciousness or who cannot actively express discomfort. Furthermore, by constructing an integrated safety mechanism for risk grading and early warning, and tiered protection strategies for emergency response, the system deeply links early warning signals with bedside movement control. Tiered protection actions are automatically initiated at the initial stage of risk, and if necessary, the patient can quickly return to a supine position and lock the position without manual intervention. This significantly shortens emergency response delays and maximizes the safety of critically ill stroke patients during postural training. This solves problems such as delayed control methods, high false alarm rates, and passive risk responses.

[0097] Figure 3 is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0098] The memory 301, the processor 302, and the computer program stored on the memory 301 and capable of running on the processor 302.

[0099] When the processor 302 executes the program, it implements the non-electric variable intelligent feedback control method for the electric standing bed for severe stroke rehabilitation provided in the above embodiments.

[0100] Furthermore, electronic devices also include:

[0101] Communication interface 303 is used for communication between memory 301 and processor 302.

[0102] The memory 301 is used to store computer programs that can run on the processor 302.

[0103] The memory 301 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0104] If the memory 301, processor 302, and communication interface 303 are implemented independently, then the communication interface 303, memory 301, and processor 302 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0105] Optionally, in a specific implementation, if the memory 301, processor 302, and communication interface 303 are integrated on a single chip, then the memory 301, processor 302, and communication interface 303 can communicate with each other through an internal interface.

[0106] Processor 302 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0107] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for intelligent feedback control of non-electric variables in an electric standing bed for severe stroke rehabilitation.

[0108] This application also provides a computer program product, which stores a computer program that, when executed by a processor, implements the above-mentioned non-electric variable intelligent feedback control method for an electric standing bed for severe stroke rehabilitation.

[0109] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0110] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0111] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0112] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0113] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A non-electric variable intelligent feedback control system for an electric standing bed for severe stroke rehabilitation, characterized in that: include: The module includes a signal sensing module, a signal analysis module, a standing control module, and a safety protection and emergency response module. The signal sensing module is used to collect physiological variable data through a non-invasive sensor array. The physiological variable data includes the patient's body surface pressure distribution, respiratory rhythm, and muscle tension. The signal analysis module is used to extract features from the physiological variable data to obtain multidimensional feature vectors, perform recursive analysis on the multidimensional feature vectors based on a time-series prediction model, generate a physiological state trend curve, and analyze the patient's current physiological reserve threshold and expected imbalance probability according to a preset postural tolerance model for severe stroke. The signal analysis module includes: a feature extraction unit, a physiological state trend curve generation unit, and a postural tolerance estimation unit. The feature extraction unit receives physiological variable data, performs clinical feature mining and nonlinear dynamic feature extraction to obtain multidimensional feature vectors. The physiological state trend curve generation unit builds a time-series prediction model based on an LSTM-RNN network, inputs the multidimensional feature vectors into the trained time-series prediction model, outputs a predicted value sequence of each physiological indicator within a preset future time window, and generates a physiological state trend curve based on the predicted value sequence. The postural tolerance estimation unit builds a postural tolerance model for severe stroke, inputs the multidimensional feature vectors and the physiological state trend curve into the trained postural tolerance model for severe stroke for personalized matching and correlation inference, and outputs the patient's current physiological reserve threshold and expected imbalance probability. The construction and operation of the postural tolerance model for severe stroke includes the following steps: collecting and structuring multi-source medical knowledge in the field of severe stroke, constructing a knowledge graph for postural tolerance assessment, wherein the nodes of the knowledge graph include patient static attributes, stroke lesion characteristics, rehabilitation stage, and postural response, and the edges of the knowledge graph represent the medical semantic relationships between nodes; using a deep neural network with a multi-head attention mechanism to build a feature learner, mining deep feature representations related to postural tolerance from the multi-dimensional feature vectors and the physiological state trend curves to obtain deep feature vectors; matching the knowledge graph with the current patient's individual characteristics and real-time rehabilitation stage, aligning the deep feature vectors with the matched knowledge graph entities across modalities, aggregating the first-order and multi-hop neighbor information of entities through a graph attention network to generate a knowledge-enhanced patient representation; inputting the knowledge-enhanced patient representation into a multi-task learning output layer, jointly optimizing the physiological reserve threshold regression task and the expected imbalance probability classification task by sharing hidden layers and task-specific branches, wherein the physiological reserve threshold regression task outputs the patient's current physiological reserve threshold, and the expected imbalance probability classification task outputs the patient's current expected imbalance probability; The standing control module is used to adaptively adjust the standing angle and angular velocity of the electric standing bed according to the physiological reserve threshold and the expected imbalance probability. When the expected imbalance probability exceeds the safe range or the physiological reserve threshold is insufficient, the safety protection is triggered. The safety protection and emergency response module is used to generate multi-level hierarchical early warning signals when the physiological state trend curve reaches a preset risk threshold or when the standing control module triggers safety protection, and to generate corresponding hierarchical protection strategies based on the multi-level hierarchical early warning signals.

2. The non-electric variable intelligent feedback control system for the electric standing bed for severe stroke rehabilitation according to claim 1, characterized in that, The signal sensing module includes: a body surface pressure sensing unit, a respiratory rhythm sensing unit, a muscle tension sensing unit, and a data preprocessing unit, wherein... The body surface pressure sensing unit is used to collect static pressure distribution images and dynamic pressure change time sequences of the patient in contact with the bed surface through a distributed high-density flexible piezoresistive sensor array integrated in the weight-bearing area of ​​the mattress, and to extract the body position center of gravity offset and body weight-bearing symmetry index in real time through a pressure center trajectory calculation algorithm. The respiratory rhythm sensing unit is used to capture mechanical vibration signals caused by respiratory movements in the patient's chest and abdomen through a vibration sensor array embedded in the corresponding chest and abdomen area of ​​the mattress. After removing interference through an adaptive filtering algorithm, the respiratory rate, inspiratory / expiratory time ratio and relative tidal volume change trends are extracted. The muscle tension sensing unit is used to collect electromyographic signals during muscle activity by using elastic strain gauge non-electric sensing components attached to the elbow of the patient's upper limb and the main muscle groups of the lower limb. The root mean square value, median frequency and electromyographic integral value are extracted by time-frequency domain analysis. The data preprocessing unit is used to perform spatiotemporal alignment and missing value filling on the data collected by the above-mentioned sensing unit, and then integrate the data to obtain physiological variable data.

3. The non-electric variable intelligent feedback control system for the electric standing bed for severe stroke rehabilitation according to claim 1, characterized in that, The standing control module includes: an adaptive standing planning unit and a safe collaborative execution unit, wherein... The adaptive standing planning unit is used to receive the physiological reserve threshold and the expected imbalance probability in real time, and dynamically obtain the optimal standing movement parameters through a multi-objective optimization algorithm according to the preset rehabilitation training target curve. The standing movement parameters include target angle, angular velocity curve, and pause point. The safety collaborative execution unit is used to convert the standing motion parameters into motor control commands to drive the electric standing bed to move, and simultaneously monitor the physiological reserve threshold and the expected imbalance probability. When the expected imbalance probability exceeds the safe range or the physiological reserve threshold is insufficient, it switches to the safety protection mode. The safety protection mode restores the bed to the supine position at the fastest safe speed and keeps the position locked until it is manually released.

4. The non-electric variable intelligent feedback control system for the electric standing bed for severe stroke rehabilitation according to claim 1, characterized in that, The security protection and emergency response module includes: a risk classification and early warning unit, a classification protection strategy generation unit, and an emergency response execution unit, wherein... The risk classification and early warning unit is used to receive physiological state trend curves, physiological reserve thresholds, and expected imbalance probabilities, and divide risk events into three levels and generate early warning signals corresponding to the levels based on a preset multi-dimensional risk quantification assessment matrix. The graded protection strategy generation unit is used to receive the early warning signal and generate a graded protection strategy that matches the current risk level and individual patient characteristics through the built-in rule engine decision model. The emergency response execution unit is used to convert the graded protection strategy into an executable instruction sequence, and drive the bed mechanism and auxiliary devices to perform protective actions according to the instruction sequence.

5. A method for applying the non-electric variable intelligent feedback control system of the electric standing bed for severe stroke rehabilitation according to any one of claims 1-4, characterized in that, Includes the following steps: Physiological variable data are collected using a non-invasive sensor array, including patient body surface pressure distribution, respiratory rhythm, and muscle tone. The physiological variable data are subjected to feature extraction to obtain multidimensional feature vectors. Based on the time series prediction model, the multidimensional feature vectors are recursively analyzed to generate physiological state trend curves. The patient's current physiological reserve threshold and expected imbalance probability are analyzed according to the preset postural tolerance model for severe stroke. The standing angle and angular velocity of the electric standing bed are adaptively adjusted according to the physiological reserve threshold and the expected imbalance probability. When the expected imbalance probability exceeds the safe range or the physiological reserve threshold is insufficient, the safety protection is triggered. When the physiological state trend curve reaches a preset risk threshold or the standing control module triggers safety protection, a multi-level hierarchical early warning signal is generated, and a corresponding hierarchical protection strategy is generated based on the multi-level hierarchical early warning signal.

6. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the non-electric variable intelligent feedback control method for the electric standing bed for severe stroke rehabilitation as described in claim 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the non-electric variable intelligent feedback control method for the electric standing bed for severe stroke rehabilitation as described in claim 5.

8. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed, they implement the non-electric variable intelligent feedback control method for the electric standing bed for severe stroke rehabilitation as described in claim 5.

Images