Intelligent nursing system for nursing pressure sores of cerebral infarction patients and back lifting optimization method

Abstract

The invention relates to the technical field of pressure sore nursing, discloses an intelligent nursing system for pressure sore nursing of a patient suffering from cerebral infarction and a back raising optimization method, and is particularly used for solving the problem that the risk of pressure sore deterioration is caused due to the fact that the patient suffering from cerebral infarction cannot sense local pressure and abnormal shearing force in the back raising process in time due to feeling and dyskinesia. The system comprises a data acquisition module, a feature extraction module, a feature fusion module, a feature dimension reduction module, a pressure sore risk prediction module, a feedback control module and a back-up execution module, and the method is based on a target function, and dynamically adjusts back-up execution parameters in combination with acquired data and pressure sore risk scores and grades obtained through analysis. According to the method, multi-modal data are collected in real time, the body posture, motion trail, local stress, shearing force and pressure of the patient in the back raising process are comprehensively considered, closed-loop control is optimized through an objective function, the back raising speed, angle and supporting force are dynamically adjusted, and pressure sore aggravation or secondary injury of the patient after cerebral infarction in the back raising process is prevented.

Description

Technical Field

[0001] The present invention relates to the technical field of pressure ulcer care, and more specifically, to an intelligent care system for pressure ulcer care of patients with sequelae of cerebral infarction and a back-lifting optimization method therefor. Background Art

[0002] It is pointed out in the publicly available CNKI literature "Research on the Compensation Mechanism for the Back Slipping of Nursing Beds" that the mechanisms for the formation of pressure ulcers include blood circulation disorders, tissue hypoxia, and injury stress responses. When patients move on nursing beds or sit in wheelchairs, the skin and body tissues are constantly subjected to reverse resistance friction from the bed sheets and the surfaces of wheelchairs. The frictional force will enhance the shearing force acting on the skin and body tissues. When the skin adheres to the bed surface, the movement of the deep body tissues will cause greater shear stress, resulting in blood vessel damage and tissue deformation, and increasing the risk of infection. Especially when the patient is being lifted up, with the upper body being elevated on the bed, the body sliding and object dislocation generate both frictional force and pressure, further increasing the risk of pressure ulcer formation. The Chinese invention patent with the application number 2024101696659 discloses an information management system and method for patient care safety, which performs feature extraction and encoding on the patient activity monitoring video within a predetermined time period, the captured images of specific parts of the patient's skin, and the patient's clinical data based on artificial intelligence technology in the field of deep learning to obtain a risk classification result for whether the patient has a pressure ulcer, and intelligently and real-time judges whether the patient has a risk of pressure ulcer, which helps prevent the development and further deterioration of pressure ulcer ulcers and improves the safety of patient care. Although this system and method perform feature extraction and encoding based on the patient's clinical data, thereby providing a real-time monitoring and feedback mechanism, it does not consider the sensory and motor impairments of patients with sequelae of cerebral infarction, nor does it timely monitor the shearing force and pressure on the skin and tissues during the process of the patient getting up from the nursing equipment after suffering from a pressure ulcer to evaluate the risk of pressure ulcer and secondary injury caused to the patient by the current back-lifting operation parameters, and it is unable to analyze the intelligent adjustment parameters for the back-lifting execution mechanism in the nursing equipment based on the currently obtained patient feature data. Summary of the Invention

[0003] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent care system for pressure ulcer care of patients with sequelae of cerebral infarction and a back-lifting optimization method therefor. By real-time collecting multi-modal data and comprehensively considering the body posture, movement trajectory, local force, shearing force, and pressure during the back-lifting process of the patient, and using the objective function to optimize the closed-loop control, the back-lifting speed, angle, and support strength are dynamically adjusted to prevent the pressure ulcer of the patient with sequelae of cerebral infarction from worsening or secondary injury during the back-lifting process.

[0004] In the case of a cerebroinfarction sequela patient already suffering from pressure ulcers, due to the sensory and motor disorders caused by the damaged nerve function, the patient is unable to promptly perceive local pain and abnormal force, and in addition, the skin and deep tissues at the pressure ulcer site are already in a fragile state. Traditional pressure ulcer prevention methods mainly focus on regulating the local force distribution when pressure ulcers have not occurred, and it is difficult to cope with the severe shear force, friction, pressure concentration generated during the back-lifting process due to the change of the patient's body position, as well as the resulting local blood circulation disorder, tissue hypoxia and injury stress response, thus extremely likely to induce the deterioration of pressure ulcers and secondary injuries.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] An intelligent nursing system for pressure ulcer care of cerebroinfarction sequela patients, including a data acquisition module, a feature extraction module, a feature fusion module, a feature dimension reduction module, and a pressure ulcer risk prediction module. The data extraction module is used to obtain the patient's activity video, skin photographed images, clinical data and real-time sensing data during the back-lifting within the observation time window. The feature extraction module is used to extract body posture features, skin features, clinical data features, and real-time back-lifting features, and focus on marking and processing the relevant index data of sensory and motor disorders of cerebroinfarction sequela patients in the clinical data. The feature fusion module is used to fuse the patient's body posture features, skin features, clinical data features, and real-time back-lifting features, and fuse the features using the high-dimensional space unit manifold sub-dimension super-convex correlation measurement method. At the same time, the clinical data feature matrix is extracted through a context encoder based on a transformer, and the indexes of cerebroinfarction sequela are encoded key points to generate a patient pressure ulcer prediction feature matrix. The pressure ulcer risk prediction module, based on the pressure ulcer prediction feature matrix, combines the shear force, pressure, local blood circulation disorder, tissue hypoxia and injury stress response indexes suffered by the skin and tissues during the back-lifting process obtained from the analysis of the back-lifting feature data, constructs an improved deep learning risk assessment model, and outputs the risk score and grade of pressure ulcer secondary injury. The pressure ulcer risk prediction module is connected to a closed-loop control module, and the closed-loop control module is connected to a back-lifting actuator. The closed-loop control module forms a closed-loop control according to the risk score and grade and real-time monitoring data, dynamically adjusts the back-lifting speed, lifting angle and trajectory, support strength and position of the back-lifting actuator according to the risk assessment result, and automatically triggers an alarm when the risk index continuously exceeds the safety threshold.

[0007] As a further solution of the present invention, in the data extraction module, the real-time sensing data of the back-lifting of the patients with sequelae of cerebral infarction includes but is not limited to the instantaneous pressure value of the contact area between the skin and the back-lifting actuator, the lateral and tangential forces received by the skin and tissues during the back-lifting process, the temperature and humidity of the contact part and their changes, the body posture changes, the lifting angle, the displacement amount, the acceleration, and the vibration data. Among them, the back-lifting movement trajectory is automatically generated according to the body posture changes, the lifting angle, and the displacement amount data, and the action amplitude evaluation index is output. The action smoothness evaluation index and the action impact index are automatically obtained according to the acceleration and vibration data.

[0008] As a further solution of the present invention, in the data extraction module, the body posture changes, the lifting angle, and the displacement amount data of the patient are synchronized according to the time stamp. Taking the initial state of the patient as the reference origin, the continuous position coordinates of the patient in the three-dimensional space are constructed by using the collected displacement amount and angle data. The Kalman filter is used to smooth the discrete sampling points to generate the back-lifting movement trajectory, which describes the curve of the position of the key parts of the patient changing with time during the back-lifting process. The normalized value of the absolute value of the position change of the key parts of the patient from the starting state to the completion of the back-lifting action is used as the action amplitude evaluation index. The collected acceleration and vibration data are subjected to low-pass filtering and noise reduction to obtain the root mean square value, the standard deviation, and the average value of the acceleration and vibration data. The normalized value of the product of the standard deviation and the root mean square of the acceleration data is used as the smoothness evaluation index. The instantaneous peak value and the sharp change point are automatically detected in the acceleration data, and their amplitudes and durations are recorded. The derivative of the acceleration data is obtained. The normalized value of the weighted sum of the instantaneous peak amplitude, the frequency of the acceleration detection, and the derivative of the acceleration data is used as the impact evaluation index.

[0009] As a further solution of the present invention, in the feature extraction module, the index data related to the sensory disturbance and motor disturbance of the patients with sequelae of cerebral infarction that are key-marked and processed in the clinical data features include but are not limited to the nerve function score, the muscle strength test result, the sensory hypoesthesia and abnormal feedback data, the limb coordination evaluation data, and the reflex state data.

[0010] As a further solution of the present invention, the process of feature fusion by the feature fusion module includes:

[0011] Step 1, data preprocessing and standardization: Denoise, normalize, and standardize the body posture, skin, clinical data, real-time sensing data of back-lifting, and back-lifting evaluation index. Align different sampling points of the real-time sensing data of back-lifting and the body posture data based on the time stamp;

[0012] Step 2, Key Encoding of Clinical Data: Input the clinical data into the set embedding layer, map discrete and continuous numerical and categorical data into a specified high-dimensional feature space, process the clinical data sequence using a Transformer-based context encoder, capture the correlations within the data through the self-attention mechanism, and assign higher attention weights to the indicators and data related to sensory disorders and motor disorders in patients with sequelae of cerebral infarction, generating a clinical feature matrix;

[0013] Step 3, Mapping and Projection of Each Maternal-Fetal Feature: Use a predefined non-linear mapping function to project the body posture features, skin features, clinical data features processed in Step 2, and real-time back-lifting features into the same set target high-dimensional feature space respectively;

[0014] Step 4, Constructing a High-Dimensional Manifold Model: Take the mapped features as sample points in the target high-dimensional space, use the manifold learning method of locally linear embedding to explore the distribution structure of the sample points on the manifold in the set low-dimensional space, and extract the low-dimensional representation of each feature in the manifold;

[0015] Step 5, Hyperconvex Correlation Measurement and Weight Optimization: Use the data obtained in Step 4 to construct a preliminary feature correlation matrix, and use the hyperconvex optimization method to optimize the correlation matrix under the constraint of the globally unique optimal solution to solve for the optimal combination weights;

[0016] Step 6, Multimodal Feature Weighted Fusion and Output: According to the optimal combination weights obtained in Step 5, perform weighted summation on each feature after target high-dimensional mapping, and perform normalization, noise reduction, and format unification processing on the result of the weighted summation, and integrate it into a patient pressure ulcer prediction feature matrix and output.

[0017] As a further solution of the present invention, the process of the pressure ulcer risk prediction module predicting the risk score and level based on the input data includes:

[0018] Step 1, Data Input: Use historical nursing data and clinical outcomes to divide the data samples into a training set, a validation set, and a test set, and construct pressure ulcer secondary injury risk labels according to the clinical evaluation results as the target of supervised learning;

[0019] Step 2, Constructing Model Input: Concatenate the patient pressure ulcer prediction feature matrix and the real-time back-lifting indicators in a predetermined format to form a multimodal input vector, use an additional time series encoding layer to encode the continuous sensing data during the back-lifting process to make it work in coordination with the static feature vector, use the embedding layer to convert discrete and categorical features into high-dimensional vectors, and then concatenate them with the continuous features;

[0020] Step 3, Design the model architecture: Input the static information from the patient pressure ulcer prediction feature matrix in the multi-modal input vector into the static branch, and input the real-time sensing data during the backrest raising process into the dynamic branch. The dynamic branch uses an LSTM model. Inside each branch, several convolutional layers and fully connected layers are used to extract high-level features. The self-attention mechanism is used to integrate the features of the static branch and the dynamic branch to obtain a joint feature representation. The joint feature representation is input into several layers of fully connected networks, and after being processed by a non-linear activation function, discriminative depth features are extracted. Its output layer includes a continuous risk score output and a risk level classification output, which respectively output the risk score and risk level of secondary injury to the pressure ulcer during the backrest raising process. The mean square error loss function is used for the continuous risk score, and the cross-entropy loss function is used for the risk level classification part. The weight adjustment for class imbalance is added to ensure sensitivity to the special risks of patients with sequelae of cerebral infarction. The multi-task learning strategy is adopted to sum the two parts of the loss weighted to form a total loss function. The model is supervised and learned using historical data. The Adam optimizer is used to adjust the model weights. The performance of the model is monitored using the validation set. The early stopping strategy is adopted and the learning rate is adjusted regularly. Hyperparameters such as the number of network layers, the number of nodes, the fusion method, and the learning rate are adjusted, and the best matching model configuration is obtained through cross-validation;

[0021] Step 4, Model evaluation and risk output: Evaluate the regression accuracy and classification accuracy of the model on the test set, calculate the mean square error, accuracy, and recall rate, output the continuous risk score and the risk level, and compare them with the preset safety threshold to output real-time warning information.

[0022] As a further solution of the present invention, the process by which the closed-loop control module obtains the control instruction for the backrest raising actuator according to the input data includes:

[0023] Step 1, Safety threshold comparison: Compare the risk prediction result with the preset safety threshold, and combine the real-time sensing data of the backrest raising to determine whether there are potential risks in the current backrest raising process;

[0024] Step 2, Error signal generation and control strategy determination: Generate an error signal based on the deviation between the actually measured real-time data of the backrest raising and the preset ideal safety state. Based on the PID control algorithm, calculate the amplitude and direction of the parameters that need to be adjusted according to the error signal, generate a control instruction for dynamically modifying the backrest raising action parameters of the backrest raising actuator, and transmit the instruction to the backrest raising actuator;

[0025] Step 3, Closed-loop control feedback and alarm mechanism: Continuously collect the adjusted real-time data of the backrest raising, compare the new real-time data of the backrest raising with the safety standard to form a closed-loop feedback loop. If the risk assessment index still continuously exceeds the safety threshold after adjustment, the alarm mechanism is automatically triggered to notify the medical staff to intervene.

[0026] Back-lifting optimization method for pressure ulcer care of patients with sequelae of cerebral infarction, applying the above intelligent nursing system for pressure ulcer care of patients with sequelae of cerebral infarction, including the following steps:

[0027] Step 1: Real-time data collection and modeling: Using a camera in cooperation with a bone tracking algorithm, and real-time displacement and angle sensors, collect data on the patient's body posture, back-lifting angle, and displacement. Taking the patient's initial static state as the reference origin, map the discrete displacement and angle data to a three-dimensional coordinate system to construct a continuous back-lifting motion trajectory;

[0028] Step 2: Adaptive optimization of back-lifting angle and motion trajectory: Based on evaluation indicators of local pressure, shear force, and motion smoothness during the back-lifting process, construct an objective function. On the basis of a preset ideal trajectory, use the dynamic programming algorithm to update the motion path in real time. When it is detected that the rising rate and amplitude of the local shear force and pressure exceed the preset threshold, adjust the motion trajectory through the objective function to avoid the risk area and reduce the back-lifting support force. At the same time, calculate the ideal lifting angle according to the changes in the patient's body posture and motion trajectory, and use a multi-degree-of-freedom servo motor to fine-tune the back-lifting angle;

[0029] Step 3: Dynamic adjustment of support strength and support pad position: Real-time collect pressure data in each area to form a pressure distribution map. Through image processing technology and data analysis algorithms, detect pressure concentration areas and local outliers, and calculate the actual load of each support pad. After setting the target pressure distribution, use a prediction model based on neural network to calculate the ideal support output strength. Compare the actual force at the current support point with the target value to determine the required adjustment amount of the support strength. At the same time, use the fine-tuning displacement technology combined with real-time pressure analysis data to calculate the optimal support position, and dynamically adjust the support position, and real-time feedback the adjusted support strength and position data to form a closed-loop control to ensure continuous optimal back-lifting support state.

[0030] As a further solution of the present invention, step 2 simultaneously considers trajectory deviation, local shear force, local pressure, and motion smoothness to construct an objective function, and the objective function is as follows:

[0031]

[0032] In the formula: J is the objective function value, equal to the total cost accumulated at each time step within the planning time domain, t is the index of the discrete time step, T is the total number of time steps of the plan, ω1 is the trajectory deviation weight, set by expert experience, ω2 is the force penalty weight, determined based on clinical standards, ω3 is the weight of control smoothness, used to punish the drastic change of control instructions between consecutive time steps to ensure smooth motion, s(t) is the current back-lifting state vector of the patient at time t, s ref(t) is the ideal back-lifting state vector preset at time t, ||s(t - s ref (t)|| 2 is the Euclidean square between the actual state and the ideal reference state used to quantify the deviation of the back-lifting motion trajectory. P(t) is the shear force and pressure values exerted on the patient's local skin and tissue at time t. Penalty(P(t)) is the force penalty function, which increases the penalty when P(t) exceeds the preset safety threshold (Penalty(P(t)) = α·(P(t) - T P )) 2 , otherwise P(t) = 0. α is the penalty coefficient, determined based on clinical safety standards. T P is the preset safety threshold for the shear force and pressure values exerted on the skin and tissue. u(t) is the input sent to the back-lifting actuator at time t, obtained by closed-loop control. u(t + 1) is the control input vector sent to the back-lifting actuator at the discrete time step (t + 1).

[0033] As a further solution of the present invention, in the objective function of step two, the trigger thresholds for pressure and shear force are 20% lower than the normal thresholds.

[0034] Technical effects of the intelligent nursing system and back-lifting optimization method for pressure ulcer care of patients with sequelae of cerebral infarction of the present invention:

[0035] The present invention constructs a continuous three-dimensional back-lifting motion trajectory by using a camera and a bone tracking algorithm in cooperation with multi-modal sensing data of real-time displacement, angle, pressure, shear force, and acceleration. On this basis, an objective function is constructed based on dynamic programming or model predictive control to adaptively adjust the motion trajectory and finely tune the lifting angle of the area with excessive local force. At the same time, the support strength and support position are dynamically adjusted in combination with the pressure distribution map of the support pad and the neural network prediction model to achieve precise real-time control of the back-lifting process, ensuring that when local risk indicators are detected to exceed the preset safety threshold, the local load of the patient can be quickly and smoothly reduced, thereby effectively preventing the further deterioration of pressure ulcers and improving the quality of nursing. Brief Description of the Drawings

[0036] Figure 1 is the system block diagram of the intelligent nursing system for pressure ulcer care of patients with sequelae of cerebral infarction of the present invention;

[0037] Figure 2 is the flowchart of the back-lifting optimization method for pressure ulcer care of patients with sequelae of cerebral infarction of the present invention;

[0038] Figure 3 is the screenshot of the data monitoring interface of the system proposed by the present invention. Detailed Embodiments

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0040] Embodiment 1

[0041] As Figure 1 shown, the intelligent nursing system for pressure ulcer care of patients with sequelae of cerebral infarction proposed by the present invention includes a data acquisition module, a feature extraction module, a feature fusion module, a feature dimensionality reduction module, and a pressure ulcer risk prediction module. The data extraction module is used to acquire the patient's activity video, skin photograph, clinical data, and real-time sensing data of back lifting within the observation time window. The feature extraction module is used to extract body posture features, skin features, clinical data features, and real-time back lifting features, and to focus on annotating and processing the relevant index data of sensory impairment and motor impairment of patients with sequelae of cerebral infarction in the clinical data. The feature fusion module is used to fuse the patient's body posture features, skin features, clinical data features, and real-time back lifting features, and to fuse the features using the high-dimensional space unit manifold sub-dimension hyperconvex correlation measurement method. At the same time, the clinical data feature matrix is extracted through the context encoder based on the transformer, and the indexes of sequelae of cerebral infarction are encoded emphatically to generate the patient's pressure ulcer prediction feature matrix. The pressure ulcer risk prediction module is based on the pressure ulcer prediction feature matrix, and combines the shear force, pressure, local blood circulation disorder, tissue hypoxia, and injury stress response indexes of the skin and tissues during the back lifting process obtained from the analysis of the back lifting feature data, constructs an improved deep learning risk assessment model, and outputs the risk score and grade of secondary pressure ulcer injury. The pressure ulcer risk prediction module is connected to a closed-loop control module, and the closed-loop control module is connected to a back lifting actuator. The closed-loop control module forms a closed-loop control according to the risk score and grade and the real-time monitoring data, dynamically adjusts the back lifting speed, lifting angle and trajectory, support strength and position of the back lifting actuator according to the risk assessment result, and automatically triggers an alarm when the risk index continuously exceeds the safety threshold.

[0042] The data acquisition module is connected to the feature extraction module, the feature extraction module is connected to the feature fusion module, the feature fusion module is connected to the pressure ulcer risk prediction module, the pressure ulcer risk prediction module is connected to the closed-loop control module, and the closed-loop control module is connected to the back lifting actuator. The back lifting actuator is connected to the data acquisition module to form a closed-loop feedback loop.

[0043] It should be noted that in the data extraction module, the real-time sensing data of the back-lifting of stroke sequela patients includes but is not limited to the instantaneous pressure value of the skin contact area with the back-lifting actuator, the lateral and tangential forces (shearing forces) exerted on the skin and tissues during back-lifting, the temperature and humidity of the contact area and their changes, body posture changes, lifting angle, displacement, acceleration, and vibration data. Among them, the back-lifting movement trajectory is automatically generated based on the body posture changes, lifting angle, and displacement data, and the action amplitude evaluation index is output. The action smoothness evaluation index and action impact index are automatically obtained based on the acceleration and vibration data.

[0044] The instantaneous pressure value of the skin contact area with the back-lifting actuator is directly related to the local blood circulation and tissue oxygen supply. Real-time monitoring helps prevent the deterioration of pressure ulcers caused by excessive pressure. Secondly, the lateral and tangential forces (shearing forces) exerted on the skin and tissues during back-lifting are the direct manifestations of friction and relative movement. High shearing forces are likely to cause microvascular damage and tissue deformation. Monitoring this index can guide the adjustment of the movement trajectory to reduce friction damage. At the same time, the temperature and humidity of the contact area and their changes reflect the local microenvironmental conditions. Abnormal temperature and humidity will weaken the skin barrier function, thus increasing the risk of infection. In addition, the body posture changes, lifting angle, and displacement data are used to automatically generate the back-lifting movement trajectory and output the action amplitude evaluation index. These data can accurately describe the movement amplitude and path of the patient during back-lifting, ensuring smooth movement and compliance with safety requirements. The acceleration and vibration data reveal the dynamic response of the movement and can automatically obtain the action smoothness and impact indexes to prevent secondary injuries caused by sudden acceleration or violent vibration. Generally speaking, by comprehensively collecting and analyzing these multi-modal indexes, the intelligent nursing system can real-time monitor and evaluate the local force state and movement dynamics of the patient during back-lifting, and dynamically adjust the back-lifting speed, angle, trajectory, as well as the support strength and position based on the closed-loop control strategy, ensuring personalized, precise, and safe back-lifting optimization in the case where the patient cannot subjectively feedback discomfort due to sensory impairment, thus effectively reducing the risk of further deterioration of pressure ulcers and secondary injuries and improving the overall nursing quality.

[0045] It should be noted that in the data extraction module, the data of the patient's body posture changes, lifting angle, and displacement are synchronized according to the time stamp. Taking the patient's initial state as the reference origin, the collected displacement and angle data are used to construct the continuous position coordinates of the patient in the three-dimensional space. The Kalman filter is used to smooth the discrete sampling points to generate the back-lifting motion trajectory, which describes the curve of the position of the key parts of the patient changing with time during the back-lifting process. The normalized value of the absolute value of the position change of the key parts of the patient from the starting state to the completion of the back-lifting action is used as the action amplitude evaluation index. The collected acceleration and vibration data are subjected to low-pass filtering and noise reduction to obtain the root mean square value, standard deviation, and average value of the acceleration and vibration data. The normalized value of the product of the standard deviation and the root mean square of the acceleration data is used as the stability evaluation index. The instantaneous peak value and the sharp change point are automatically detected in the acceleration data, and their amplitudes and durations are recorded. The derivative of the acceleration data is obtained, and the normalized value of the weighted sum of the instantaneous peak amplitude, frequency, and the derivative of the acceleration data detected is used as the impact evaluation index.

[0046] By synchronizing the data of the patient's body posture changes, lifting angle, and displacement according to the time stamp, constructing a continuous three-dimensional motion trajectory with the initial static state as the reference origin, using the Kalman filter to smooth the discrete sampling points to generate a curve describing the position of the key parts changing with time, then using the normalized absolute displacement value to evaluate the action amplitude, and at the same time performing low-pass filtering and noise reduction processing on the acceleration and vibration data to obtain their root mean square, standard deviation, and average value, and using the normalized value of the product of the acceleration standard deviation and the root mean square as the stability evaluation index, and by automatically detecting the instantaneous peak value, sharp change point of the acceleration, recording their amplitudes, frequencies, and the weighted sum of the derivatives normalized as the impact evaluation index, this technical solution can quantitatively measure the dynamic force, motion amplitude, stability, and impact effect of the patient during the back-lifting process in real time and accurately, thereby providing accurate data support for closed-loop control and effectively guiding the back-lifting optimization operation to reduce the risk of secondary pressure ulcer injury.

[0047] It should be noted that in the feature extraction module, the data of the related indicators of the sensory impairment and motor impairment of the patients with sequelae of cerebral infarction, which are key-marked and processed in the clinical data features, include but are not limited to the neurological function score, muscle strength test results, sensory hypoesthesia and abnormal feedback data, limb coordination evaluation data, and reflex state data.

[0048] The reason for highlighting and processing these clinical data indicators (including neurological function scores, muscle strength test results, sensory hypoesthesia and abnormal feedback, limb coordination assessment, and reflex status data) is that due to damage to the central nervous system, patients with sequelae of cerebral infarction often exhibit sensory disorders and reduced motor function, making it difficult for them to autonomously perceive changes in local pressure and shear force during the back-lifting process, thus easily triggering or exacerbating pressure ulcers and secondary injuries. Therefore, by closely monitoring these indicators, the intelligent nursing system can accurately evaluate the patient's nerve response and motor ability, and then guide the back-lifting actuator to intelligently adjust the back-lifting speed, lifting angle, movement trajectory, and support strength according to the patient's actual functional status, ensuring that the risk of local force, friction, and impact is minimized during the back-lifting process, effectively preventing the further deterioration of pressure ulcers, and providing objective data support for medical staff.

[0049] It should be noted that the process of feature fusion by the feature fusion module includes:

[0050] Step 1, data preprocessing and standardization: Denoise, normalize, and standardize body posture, skin, clinical data, real-time back-lifting sensor data, and back-lifting evaluation indicators, and align different sampling points of the real-time back-lifting sensor data and body posture data based on the timestamp.

[0051] Step 2, key encoding of clinical data: Input the clinical data into the set embedding layer, map discrete and continuous numerical and categorical data into a specified high-dimensional feature space, use a context encoder based on a transformer to process the clinical data sequence, capture the correlation within the data through the self-attention mechanism, and assign higher attention weights to the indicators and data related to sensory disorders and motor disorders in patients with sequelae of cerebral infarction than other data to generate a clinical feature matrix.

[0052] Step 3, mapping and projection of each maternal and fetal feature: Use a predefined non-linear mapping function to project body posture features, skin features, clinical data features processed in Step 2, and real-time back-lifting features into the same set target high-dimensional feature space respectively.

[0053] Step 4, constructing a high-dimensional manifold model: Use the mapped features as sample points in the target high-dimensional space, and use the manifold learning method of locally linear embedding to explore the distribution structure of the sample points on the manifold in the set low-dimensional space, and extract the low-dimensional representation of each feature in the manifold.

[0054] Step 5, hyperconvex correlation measurement and weight optimization: Use the data obtained in Step 4 to construct a preliminary feature correlation matrix, and use the hyperconvex optimization method to optimize the correlation matrix under the constraint of a globally unique optimal solution to solve for the best combination weights.

[0055] Step 6, Multimodal Feature Weighted Fusion and Output: According to the optimal combination weights obtained in Step 5, perform weighted summation on each feature after the target high-dimensional mapping, and perform normalization, noise reduction, and format unification processing on the result of the weighted summation, and integrate it into a patient pressure ulcer prediction feature matrix and output.

[0056] The set target high-dimensional space is 512-dimensional, which can fully capture the complex relationships among body posture, skin, clinical, and back-lifting real-time data. At the same time, considering the computational complexity, the high-dimensional features are reduced to a 120-dimensional low-dimensional space, which not only retains the discriminability and internal structure of the data but also effectively removes redundancy and noise, providing a more efficient and accurate input for the subsequent risk prediction model.

[0057] It should be noted that the process of the pressure ulcer risk prediction module predicting the risk score and level based on the input data includes:

[0058] Step 1, Data Input: Using historical nursing data and clinical outcomes, divide the data samples into a training set, a validation set, and a test set, and construct pressure ulcer secondary injury risk labels according to the clinical assessment results as the targets of supervised learning.

[0059] Step 2, Construct Model Input: Concatenate the patient pressure ulcer prediction feature matrix and the back-lifting real-time indicators in a predetermined format to form a multimodal input vector. Use an additional time series encoding layer to encode the continuous sensing data during the back-lifting process to make it work in coordination with the static feature vector. Use the embedding layer to convert discrete and categorical features into high-dimensional vectors and then concatenate them with the continuous features.

[0060] Step 3, Design the model architecture: Input the static information from the patient pressure ulcer prediction feature matrix in the multi-modal input vector into the static branch, and input the real-time sensing data during the back-lifting process into the dynamic branch. The dynamic branch uses an LSTM model. Inside each branch, several convolutional layers and fully connected layers are used to extract high-level features. The self-attention mechanism is used to integrate the features of the static branch and the dynamic branch to obtain a joint feature representation. The joint feature representation is input into several layers of fully connected networks, and after being processed by a non-linear activation function, discriminative depth features are extracted. Its output layer includes a continuous risk score output and a risk level classification output, which respectively output the risk score and risk level of secondary injury to the pressure ulcer during the back-lifting process. The mean squared error loss function is used for the continuous risk score, and the cross-entropy loss function is used for the risk level classification part. The weight adjustment for class imbalance is added to ensure the sensitivity to the special risks of patients with sequelae of cerebral infarction. The multi-task learning strategy is adopted to sum the two parts of the losses weighted to form the total loss function. The model is supervised and learned using historical data. The Adam optimizer is used to adjust the model weights, the validation set is used to monitor the model performance, the early stopping strategy is adopted and the learning rate is adjusted regularly. Hyperparameters such as the number of network layers, the number of nodes, the fusion method, and the learning rate are adjusted, and the best matching model configuration is obtained by means of cross-validation;

[0061] Step 4, Model evaluation and risk output: Evaluate the regression accuracy and classification accuracy of the model on the test set, calculate the mean squared error, accuracy, and recall rate, output the continuous risk score, as well as the risk level, and compare it with the preset safety threshold to output real-time warning information.

[0062] During this process, the mean squared error loss function and the cross-entropy loss function are used in combination with the class imbalance weight and the multi-task learning strategy to supervise and train the historical nursing data, and the hyperparameters such as the number of network layers, the number of nodes, the fusion method, and the learning rate are continuously adjusted through the Adam optimizer, the early stopping strategy, and cross-validation, so that the model can obtain high-precision regression and classification performance on the test set, ensure that the real-time risk assessment result can trigger a timely warning after being compared with the preset safety threshold, and provide accurate data support for the closed-loop control module to dynamically adjust the back-lifting actuator (including the back-lifting speed, the lifting angle and the movement trajectory, the supporting strength and position), thereby effectively improving the real-time monitoring and prevention and control capabilities of secondary injury to pressure ulcers in patients with sequelae of cerebral infarction, and significantly enhancing the safety and personalization level of the nursing process.

[0063] It should be noted that the process by which the closed-loop control module obtains the control instruction for the back-lifting actuator according to the input data includes:

[0064] Step 1, Safety threshold comparison: Compare the risk prediction result with the preset safety threshold, and combine the real-time sensing data of the back-lifting to judge whether there are potential risks in the current back-lifting process;

[0065] Step 2, Error Signal Generation and Control Strategy Determination: Generate an error signal based on the deviation between the actually measured real-time data of the back-lifting and the preset ideal safe state. Based on the PID control algorithm, calculate the amplitude and direction of the parameters to be adjusted according to the error signal, generate a control instruction for dynamically modifying the back-lifting action parameters of the back-lifting actuator, and transmit the instruction to the back-lifting actuator;

[0066] Step 3, Closed-loop Control Feedback and Alarm Mechanism: Continuously collect the adjusted real-time data of the back-lifting, compare the new real-time data of the back-lifting with the safety standard to form a closed-loop feedback loop. If the risk assessment index still continuously exceeds the safety threshold after adjustment, automatically trigger the alarm mechanism to notify the medical staff to intervene.

[0067] By comparing the risk prediction result with the preset safety threshold in real time through the closed-loop control module, generating an error signal in combination with the real-time sensing data of the back-lifting, calculating the adjustment amplitude and direction using the PID control algorithm, issuing a control instruction to dynamically modify the parameters of the back-lifting actuator, and at the same time continuously collecting the adjusted data to form a closed-loop feedback. Once the risk index continuously exceeds the standard, automatically trigger an alarm to notify the medical staff to intervene, thereby effectively reducing the risk of pressure ulcer deterioration caused by abnormal local pressure and shear force, realizing precise real-time prevention and control and dynamic optimization management of secondary injuries of pressure ulcers in patients with sequelae of cerebral infarction, ensuring that each adjustment step is accurate, efficient and stable, and constituting a safe, closed-loop and real-time responsive back-lifting optimization system.

[0068] Example 2

[0069] The difference between Example 2 and Example 1 of the present invention is that this example introduces an optimized back-lifting method for pressure ulcer care of patients with sequelae of cerebral infarction.

[0070] As Figure 2 shown, the optimized back-lifting method for pressure ulcer care of patients with sequelae of cerebral infarction proposed by the present invention includes the following steps:

[0071] Step 1: Real-time Data Acquisition and Modeling: Use a camera in cooperation with a bone tracking algorithm, and real-time displacement and angle sensors to collect data on the patient's body posture, back-lifting angle and displacement. Taking the patient's initial static state as the reference origin, map the discrete displacement and angle data to a three-dimensional coordinate system to construct a continuous back-lifting action trajectory;

[0072] Step 2: Adaptive Optimization of the Lifting Angle and Movement Trajectory of the Backrest: Based on the evaluation indicators of local pressure, shear force, and movement smoothness during the backrest lifting process, a target function is constructed. On the basis of the preset ideal trajectory, the dynamic programming algorithm is used to update the movement path in real time. When it is detected that the rising rate and amplitude of the local shear force and pressure exceed the preset threshold, the movement trajectory is adjusted through the target function to avoid the risk area and reduce the force on the backrest support. At the same time, the ideal lifting angle is calculated according to the changes in the patient's body posture and movement trajectory, and the multi-degree-of-freedom servo motor is used to fine-tune the lifting angle of the backrest;

[0073] Step 3: Dynamic Adjustment of the Support Strength and the Position of the Support Pad: The pressure data of each area are collected in real time to form a pressure distribution map. Through image processing technology and data analysis algorithms, the pressure concentration area and local outliers are detected, and the actual load of each support pad is calculated. After setting the target pressure distribution, a prediction model based on neural network is used to calculate the ideal support output strength. The actual force on the current support point is compared with the target value to determine the required adjustment amount of the support strength. At the same time, the best support position is calculated by using the fine-tuning displacement technology combined with the real-time pressure analysis data, and the support position is dynamically adjusted. The adjusted support strength and position data are fed back in real time to form a closed-loop control to ensure continuous optimal backrest support state.

[0074] By dynamically optimizing the execution parameters of the patient's backrest lifting process, the safety of the patient is effectively improved, and the risk of secondary injury is reduced. At the same time, the mechanical parameters during the backrest lifting process are optimized to ensure more stability when the patient's backrest is lifted and reduce the probability of accidents.

[0075] It should be noted that in Step 2, the trajectory deviation, local shear force, local pressure, and movement smoothness are considered simultaneously to construct the target function. The target function is as follows:

[0076]

[0077] Where: J is the value of the target function, which is equal to the total cost accumulated at each time step within the planning time domain. t is the index of the discrete time step, T is the total number of time steps of the plan, ω1 is the trajectory deviation weight, set by expert experience, ω2 is the force penalty weight, determined based on clinical standards, ω3 is the weight of control smoothness, used to penalize the drastic change of control commands between consecutive time steps to ensure smooth movement, s(t) is the current backrest lifting state vector of the patient at time t, s ref (t) is the preset ideal backrest lifting state vector at time t, ||s(t) - s ref (t)|| 2To quantify the squared Euclidean distance between the actual state and the ideal reference state of the back-lifting motion trajectory, P(t) represents the shear force and pressure values exerted on the patient's local skin and tissue at time t, Penalty(P(t)) is the force penalty function, which increases the penalty when P(t) exceeds the preset safety threshold (Penalty(P(t)) = α·(P(t) - T P ) 2 , otherwise P(t) = 0, α is the penalty coefficient determined based on clinical safety standards, and T P is the preset safety threshold for the shear force and pressure values exerted on the skin and tissue. u(t) is the input initiated to the back-lifting actuator at time t, obtained by closed-loop control, and u(t + 1) is the control input vector initiated to the back-lifting actuator at the discrete time step (t + 1).

[0078] It should be noted that in the objective function of step two, the trigger thresholds for pressure and shear force are 20% lower than the normal thresholds.

[0079] To clearly illustrate the specific implementation process of this method, the planning horizon is set to 3 time steps, and let ω1, ω2, ω3 be 1, 2, 0.5 respectively; at t = 0, the patient's state is exactly the same as the ideal state (||s(t) - s ref (t)|| 2 = 0), the pressure P(0) = 5 is lower than the preset threshold T P = 10 (so Penalty(P(t)) = 0), and the change in the control input ||u(t + 1) - u(t)|| 2 = 0.04; at t = 1, the squared difference between the actual state and the ideal state is 0.02, and the pressure P(1) = 12 exceeds the safety threshold, so the force penalty is α·(P(t) - T P ) 2 = 2·(12 - 10) 2 = 16 (let α = 2), the squared change in the control input is 0.09, and the total cost contribution is approximately 0.02 + 16 + 0.045 = 16.065; at t = 2, the squared state deviation is 0.005, the pressure is normal, the squared control change is 0.04, the total cost is approximately 0.005 + 0 + 0.02 = 0.025, and the cumulative cost J is 16.065. By minimizing this objective function, the system can automatically adjust the back-lifting speed, motion trajectory, and lifting angle to avoid the high-force area at t = 1, achieving effective control of the local force and motion imbalance during the back-lifting process for patients with sequelae of cerebral infarction and ensuring smooth and safe movement.

[0080] Such as Figure 3As shown, based on the system of Embodiment 1 and the method of Embodiment 2, the process uses a cloud monitoring interface to monitor the data of the entire system. The monitoring interface includes patient information, real-time monitoring, risk assessment, nursing control, alarm records, and system settings task bar. Among them, at the top of the interface, the patient's name, bed number, age, and admission time are displayed, and the video monitoring and pictures of pressure ulcer skin monitoring are displayed in real time. The vital signs of the patient are collected in real time and displayed in the form of a line chart, and the heat map of the pressure distribution of each key body part is displayed in real time. Below the data display, a backrest control panel is shown, which can automatically and manually control the backrest speed and the backrest lifting angle, and supports the display of the support strength. An emergency stop button is set to prevent emergencies. When a failure or problem occurs in the backrest actuator, the backrest execution action can be stopped in time.

[0081] In summary, it can be seen from Embodiment 1 and Embodiment 2 that the present invention constructs a continuous three-dimensional backrest movement trajectory by using a camera and a bone tracking algorithm in cooperation with multi-modal sensing data of real-time displacement, angle, pressure, shear force, and acceleration. On this basis, an objective function is constructed based on dynamic programming or model predictive control to adaptively adjust the movement trajectory and finely adjust the lifting angle of the area where the local force exceeds the standard. At the same time, the support strength and support position are dynamically adjusted in combination with the pressure distribution map of the support pad and the neural network prediction model to achieve precise real-time control of the backrest process, ensure that when local risk indicators exceed the preset safety threshold, the local load of the patient can be quickly and smoothly reduced, thereby effectively preventing the further deterioration of pressure ulcers and improving the nursing quality.

[0082] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present application, and all of them should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0083] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent nursing system for pressure ulcer care of patients with sequelae of cerebral infarction, comprising a data acquisition module, a feature extraction module, a feature fusion module, a feature dimensionality reduction module, and a pressure ulcer risk prediction module, characterized in that, The data extraction module is used to obtain the patient's activity video, skin photographs, clinical data, and real-time sensing data of the back-lifting during the observation time window. The feature extraction module is used to extract postural features, skin features, clinical data features, and real-time back-lifting features, and to highlight and process the data of related indicators of sensory impairment and motor impairment in patients with sequelae of cerebral infarction in the clinical data. The feature fusion module is used to fuse the patient's postural features, skin features, clinical data features, and real-time back-lifting features, and to fuse the features using the high-dimensional space unit manifold sub-dimensional hyperconvex correlation metric method. At the same time, the clinical data feature matrix is extracted through a transformer-based context encoder, and the indicators of sequelae of cerebral infarction are encoded with emphasis to generate the patient's pressure ulcer prediction feature matrix. The pressure ulcer risk prediction module is based on the pressure ulcer prediction feature matrix, and combines the shear force, pressure, local blood circulation disorder, tissue hypoxia, and traumatic stress response indicators of the skin and tissues during the back-lifting process obtained from the analysis of the back-lifting feature data to construct an improved deep learning risk assessment model, and output the risk score and level of secondary injury of pressure ulcers. The pressure ulcer risk prediction module is connected to a closed-loop control module, and the closed-loop control module is connected to a back-lifting actuator. The closed-loop control module forms a closed-loop control according to the risk score and level and the real-time monitoring data, and dynamically adjusts the back-lifting speed, lifting angle and trajectory, support strength and position of the back-lifting actuator according to the risk assessment result, and automatically triggers an alarm when the risk indicators continuously exceed the safety threshold.

2. The intelligent nursing system for pressure ulcer care of patients with sequelae of cerebral infarction according to claim 1, wherein, In the data extraction module, the real-time sensing data of the back-lifting of patients with sequelae of cerebral infarction includes, but is not limited to, the instantaneous pressure value of the skin contact area with the back-lifting actuator, the lateral and tangential forces received by the skin and tissues during the back-lifting process, the temperature and humidity of the contact area and their changes, body posture changes, lifting angle, displacement, acceleration, and vibration data. Among them, the back-lifting movement trajectory is automatically generated according to the body posture changes, lifting angle, and displacement data, and the action amplitude evaluation index is output. The action smoothness evaluation index and action impact index are automatically obtained according to the acceleration and vibration data.

3. The intelligent nursing system for bedsore care of patients with sequelae of cerebral infarction according to claim 2, characterized in that, In the data extraction module, the body posture changes, lifting angle, and displacement data of the patient are synchronized according to the time stamp. Taking the patient's initial state as the reference origin, the continuous position coordinates of the patient in the three-dimensional space are constructed using the collected displacement and angle data. The Kalman filter is used to smooth the discrete sampling points to generate the back-lifting movement trajectory, which describes the curve of the position of the key parts of the patient changing with time during the back-lifting process. The normalized value of the absolute value of the position change of the key parts of the patient from the starting state to the completion of the back-lifting action is used as the action amplitude evaluation index. The collected acceleration and vibration data are low-pass filtered and denoised to obtain the root mean square value, standard deviation, and average value of the acceleration and vibration data. The normalized value of the product of the standard deviation and the root mean square of the acceleration data is used as the smoothness evaluation index. The instantaneous peak value and sharp change points are automatically detected in the acceleration data, and their amplitudes and durations are recorded. The derivative of the acceleration data is obtained, and the normalized value of the weighted sum of the instantaneous peak amplitude, frequency of the acceleration detection, and the derivative of the acceleration data is used as the impact evaluation index.

4. The intelligent nursing system for pressure ulcer care of stroke sequela patients according to claim 3, wherein In the feature extraction module, the data of indicators related to sensory impairment and motor impairment of patients with sequelae of cerebral infarction, which are key-marked and processed in clinical data features, include but are not limited to neurological function scores, muscle strength test results, sensory hypoesthesia and abnormal feedback data, limb coordination assessment data, and reflex status data.

5. The intelligent nursing system for pressure ulcer care of patients with sequelae of cerebral infarction according to claim 2, characterized in that The process of feature fusion by the feature fusion module includes: Step 1, data preprocessing and standardization: Denoise, normalize, and standardize body posture, skin, clinical data, real-time back-lifting sensing data, and back-lifting assessment indicators, and align different sampling points of the real-time back-lifting sensing data and body posture data based on timestamps; Step 2, key encoding of clinical data: Input clinical data into the set embedding layer, map discrete and continuous numerical and categorical data into a specified high-dimensional feature space, process the clinical data sequence using a context encoder based on a transformer, capture the correlations within the data through a self-attention mechanism, and assign higher attention weights to the indicators and data related to sensory impairment and motor impairment in patients with sequelae of cerebral infarction to generate a clinical feature matrix; Step 3, mapping and projection of each maternal and fetal feature: Use a predefined non-linear mapping function to project body posture features, skin features, clinical data features processed in Step 2, and real-time back-lifting features into the same set target high-dimensional feature space respectively; Step 4, construct a high-dimensional manifold model: Use the mapped features as sample points in the target high-dimensional space, and use the manifold learning method of locally linear embedding to explore the distribution structure of the sample points on the manifold in the set low-dimensional space, and extract the low-dimensional representation of each feature in the manifold; Step 5, hyperconvex correlation measurement and weight optimization: Use the data obtained in Step 4 to construct a preliminary feature correlation matrix, and use the hyperconvex optimization method to optimize the correlation matrix under the constraint of the globally unique optimal solution to obtain the optimal combination weights; Step 6, multi-modal feature weighted fusion and output: According to the optimal combination weights obtained in Step 5, perform weighted summation on each feature after target high-dimensional mapping, and perform normalization, denoising, and format unification processing on the result of the weighted summation, and integrate it into a patient pressure ulcer prediction feature matrix and output.

6. The intelligent nursing system for pressure ulcer care of patients with sequelae of cerebral infarction according to claim 5, characterized in that, The process of the pressure ulcer risk prediction module predicting the risk score and level based on the input data includes: Step 1, data input: Use historical nursing data and clinical results to divide the data samples into a training set, a validation set, and a test set, and construct a pressure ulcer secondary injury risk label based on the clinical assessment results as the target of supervised learning; Step 2, construct model input: Concatenate the patient pressure ulcer prediction feature matrix and the real-time back-lifting indicators in a predetermined format to form a multi-modal input vector, use an additional time series encoding layer to encode the continuous sensing data during the back-lifting process to make it work in coordination with the static feature vector, use the embedding layer to convert discrete and categorical features into high-dimensional vectors, and then concatenate them with the continuous features; Step 3, Design the model architecture: Input the static information from the patient's pressure ulcer prediction feature matrix in the multi-modal input vector into the static branch, and input the real-time sensing data during the back-lifting process into the dynamic branch. The dynamic branch uses the LSTM model. Inside each branch, several convolutional layers and fully connected layers are used to extract high-level features. The self-attention mechanism is used to integrate the features of the static branch and the dynamic branch to obtain a joint feature representation. The joint feature representation is input into several layers of fully connected networks, and after being processed by a non-linear activation function, discriminative depth features are extracted. Its output layer includes a continuous risk score output and a risk level classification output, which respectively output the risk score and risk level of secondary injury to the pressure ulcer during the back-lifting process. The mean square error loss function is used for the continuous risk score, and the cross-entropy loss function is used for the risk level classification part. The weight adjustment for class imbalance is added to ensure sensitivity to the special risks of stroke sequela patients. The multi-task learning strategy is adopted to sum the two parts of the loss weighted to form a total loss function. The model is supervised and learned using historical data. The Adam optimizer is used to adjust the model weights, the validation set is used to monitor the model performance, the early stopping strategy is adopted and the learning rate is adjusted regularly. Hyperparameters such as the number of network layers, the number of nodes, the fusion method, and the learning rate are adjusted, and the best-matched model configuration is obtained by means of cross-validation; Step 4, Model evaluation and risk output: Evaluate the regression accuracy and classification accuracy of the model on the test set, calculate the mean square error, accuracy, and recall rate, output the continuous risk score and the risk level, and compare with the preset safety threshold to output real-time warning information.

7. The intelligent nursing system for pressure ulcer care of stroke sequela patients according to claim 1, wherein The process by which the closed-loop control module obtains the control instruction for the back-lifting actuator according to the input data includes: Step 1, Safety threshold comparison: Compare the risk prediction result with the preset safety threshold, and combine the real-time back-lifting sensing data to determine whether there are potential risks in the current back-lifting process; Step 2, Error signal generation and control strategy determination: Generate an error signal based on the deviation between the actually measured real-time back-lifting data and the preset ideal safety state. Based on the PID control algorithm, calculate the amplitude and direction of the parameters that need to be adjusted according to the error signal, generate a control instruction to dynamically modify the back-lifting action parameters of the back-lifting actuator, and transmit the instruction to the back-lifting actuator; Step 3, Closed-loop control feedback and alarm mechanism: Continuously collect the adjusted real-time back-lifting data, compare the new real-time back-lifting data with the safety standard to form a closed-loop feedback loop. If the risk assessment index still continuously exceeds the safety threshold after adjustment, the alarm mechanism is automatically triggered to notify the medical staff to intervene.

8. A back-lifting optimization method for pressure ulcer care of stroke sequela patients, which applies the intelligent care system for pressure ulcer care of stroke sequela patients described in any one of claims 1-7, characterized in that, It includes the following steps: Step 1: Real-time data collection and modeling: Use a camera in cooperation with a skeleton tracking algorithm, and real-time displacement and angle sensors to collect the patient's body posture, back-lifting angle, and displacement data. Taking the patient's initial static state as the reference origin, map the discrete displacement and angle data to a three-dimensional coordinate system to construct a continuous back-lifting action trajectory; Step 2: Adaptive optimization of the back-lifting angle and movement trajectory: Based on the evaluation indexes of local pressure, shear force and movement smoothness during the back-lifting process, a target function is constructed. On the basis of the preset ideal trajectory, the dynamic programming algorithm is used to update the movement path in real time. When it is detected that the rising rate and amplitude of the local shear force and pressure exceed the preset threshold, the movement trajectory is adjusted through the target function to avoid the risk area and reduce the force on the back-lifting support. At the same time, the ideal lifting angle is calculated according to the changes in the patient's body posture and movement trajectory, and the multi-degree-of-freedom servo motor is used to finely adjust the back-lifting angle. Step 3: Dynamic adjustment of the support strength and the position of the support pad: The pressure data of each area are collected in real time to form a pressure distribution map. Through image processing technology and data analysis algorithms, the pressure concentration area and local outliers are detected, and the actual load of each support pad is calculated. After setting the target pressure distribution, a prediction model based on neural network is used to calculate the ideal support output strength. The actual force on the current support point is compared with the target value to determine the required adjustment amount of the support strength. At the same time, the fine displacement technology is combined with the real-time pressure analysis data to calculate the optimal support position, and the support position is dynamically adjusted. The adjusted support strength and position data are fed back in real time to form a closed-loop control to ensure that the back-lifting support is continuously in the optimal state.

9. The back-lifting optimization method for pressure ulcer care of stroke sequela patients according to claim 8, characterized in that, Step 2 simultaneously considers the trajectory deviation, local shear force, local pressure and movement smoothness to construct a target function. The target function is as follows: Where: J is the objective function value, equal to the total cost accumulated at each time step within the planning time domain, t is the index of the discrete time step, T is the total number of time steps for the plan, ω1 is the trajectory deviation weight, set by expert experience, ω2 is the force penalty weight, determined based on clinical criteria, ω3 is the weight for control smoothness, used to penalize drastic changes in control commands between consecutive time steps to ensure smooth movement, s(t) is the current back-lifting state vector of the patient at time t, s ref (t) is the preset ideal back-lifting state vector at time t, ||s(t) - s ref (t)|| 2 is the squared Euclidean distance between the actual state and the ideal reference state used to quantify the back-lifting motion trajectory deviation, P(t) is the shear force and pressure values on the patient's local skin and tissue at time t, Penalty(P(t)) is the force penalty function, which increases the penalty when P(t) exceeds the preset safety threshold Penalty(P(t)) = α·(P(t) - T P ) 2 , otherwise P(t) = 0, α is the penalty coefficient, determined based on clinical safety criteria, T P is the preset safety threshold for the shear force and pressure values on the skin and tissue, u(t) is the input sent to the back-lifting actuator at time t, obtained by closed-loop control, and u(t + 1) is the control input vector sent to the back-lifting actuator at the discrete time step (t + 1).

10. The optimized back-lifting method for pressure ulcer care of stroke sequela patients according to claim 9, characterized in that, In the target function of Step 2, the trigger thresholds for pressure and shear force are 20% lower than the normal thresholds.

Images