A remote intelligent rehabilitation nursing method and system for elderly patients

Abstract

The application provides a remote intelligent rehabilitation nursing method and system for elderly patients, relates to the field of medical care information science, and determines a risk level index through vital sign parameters and clinical examination results of the elderly patients; carries out artifact suppression and frequency domain feature enhancement processing on the electroencephalogram signals of the elderly patients to obtain signal feature segments; encodes the signal feature segments according to the risk level index to further obtain a matching feature vector of a rehabilitation task of the elderly patients, carries out priority evaluation on the rehabilitation task of the elderly patients through the task matching feature vector to obtain an adaptive task; remotely controls the process of executing the adaptive task on the elderly patients to obtain a control coefficient, and carries out feedback adjustment on the nursing terminal of the elderly patients according to the control coefficient to obtain a stage adaptive task, so that the application can realize remote rehabilitation task self-adaptive matching of the elderly patients based on multi-modal feature coding to improve the individual precision of rehabilitation nursing.

Description

Technical Field

[0001] This application relates to the field of healthcare informatics, and more specifically, to a remote intelligent rehabilitation care method and system for elderly patients. Background Technology

[0002] Healthcare informatics is an emerging discipline that arises from the intersection of information technology and the healthcare field. It aims to improve the efficiency, quality, and accessibility of healthcare services by collecting, storing, processing, and sharing healthcare data through computer technology, communication technology, and data analysis.

[0003] With the increasing aging population and the continuous rise in the incidence of chronic diseases, healthcare informatics is playing an increasingly important role in scenarios such as remote health monitoring, intelligent early warning, personalized care, and rehabilitation management. Remote rehabilitation systems can provide basic training programs for elderly patients by combining a pre-set rehabilitation task library. However, current technologies still suffer from insufficient integration and utilization of multi-source heterogeneous physiological data from elderly patients, making it difficult to comprehensively depict their overall health status. Furthermore, most systems use fixed rule matching to roughly map pre-set tasks to limited patient indicators, lacking a deep adaptation mapping mechanism based on feature encoding, and thus failing to achieve real-time dynamic correction of key parameters such as task intensity and duration. Additionally, they lack the ability to continuously optimize and adjust task generation strategies based on patients' staged response results, making it difficult for the system's individualized adaptation level to continuously improve as the rehabilitation process progresses. Therefore, how to achieve adaptive matching of remote rehabilitation tasks for elderly patients based on multimodal feature encoding to improve the individualized accuracy of rehabilitation care is a challenge facing the industry. Summary of the Invention

[0004] This application provides a remote intelligent rehabilitation nursing method and system for elderly patients, which can realize adaptive matching of remote rehabilitation tasks for elderly patients based on multimodal feature encoding, so as to improve the individualized accuracy of rehabilitation nursing.

[0005] Firstly, this application provides a remote intelligent rehabilitation nursing method for elderly patients, the rehabilitation nursing method comprising the following steps: The vital signs parameters of elderly patients are collected through nursing terminals, and the risk level indicators are determined in combination with the clinical examination results of elderly patients. Electroencephalography (EEG) was used to collect EEG signals from elderly patients. The EEG signals were then processed with artifact suppression and frequency domain feature enhancement to obtain signal feature segments. The signal feature segments are feature-encoded according to the risk level index to obtain the matching feature vector of the rehabilitation task for elderly patients. The rehabilitation task of elderly patients is prioritized by the task matching feature vector to obtain the adaptive task. The process of elderly patients performing the adaptation task is remotely controlled to obtain a control coefficient. The nursing terminal of the elderly patients is then adjusted based on the control coefficient to obtain the stage adaptation task.

[0006] In this embodiment, the risk level indicators for elderly patients are determined by collecting vital sign parameters through a nursing terminal and combining them with the patients' clinical examination results. Specifically, these indicators include: A multi-parameter feature matrix was constructed based on the vital signs parameters and clinical examination results of elderly patients; The multi-parameter feature matrix is ​​preprocessed, and then a risk prediction curve for elderly patients is generated based on the preprocessed multi-parameter feature matrix. The risk level index is determined by the risk prediction curve and the preset risk threshold.

[0007] In this embodiment, the EEG signal undergoes artifact suppression and frequency domain feature enhancement processing to obtain signal feature segments, specifically including: The EEG signal was subjected to baseline drift correction and decomposition reconstruction to obtain the corrected and reconstructed signal; The corrected and reconstructed signal is converted to the frequency domain, and then the spectral distribution characteristics in the frequency domain are subjected to low-frequency suppression processing to obtain the frequency domain enhanced signal. The frequency domain enhancement signal is segmented using a time-series segmentation algorithm to obtain signal feature segments.

[0008] In this embodiment, the process of encoding the signal feature segments based on the risk level index to obtain the matching feature vector for the rehabilitation task of elderly patients specifically includes: The risk level index and the signal feature segment are time-stamped and normalized to obtain a fusion input matrix; The fused input matrix is ​​subjected to scale convolutional encoding and weight combination to obtain a rehabilitation adaptation weight map; The rehabilitation adaptation weight map is mapped and decoded to obtain the matching feature vector of the rehabilitation task for elderly patients.

[0009] In this embodiment, the rehabilitation tasks for elderly patients are prioritized using the task matching feature vector, and the resulting adaptive tasks specifically include: The matching feature vector is matched with all rehabilitation tasks in the preset task library to obtain a fit sequence. The fitness sequence is subjected to risk weighting to obtain a weighted priority sequence; A task selection discrimination matrix is ​​constructed based on the weighted priority sequence, and then the selection probability distribution of candidate tasks is obtained through the task selection discrimination matrix. The adaptation task for elderly patients is output based on the selected probability distribution.

[0010] In this embodiment, the process of elderly patients performing the adaptive task is remotely controlled based on the nursing terminal, and the control coefficient specifically includes: Determine the task performance vector and EEG signal feature vector of elderly patients when performing adaptive tasks; Based on a deep discriminant network, modality fusion is performed on the task performance vector and the EEG signal feature vector to obtain a fused representation vector; The fused representation vector is subjected to hierarchical risk discrimination to obtain the control coefficient.

[0011] In this embodiment, the nursing terminal for elderly patients is adjusted based on the control coefficient to obtain the stage adaptation task, which specifically includes: The task intensity, duration, and feedback frequency of elderly patients performing the adaptive task are adjusted in real time to obtain a task adjustment curve. Based on the changing trends of the control coefficient and the task correction curve, a fusion modeling analysis is performed to obtain a task compliance score. The nursing terminal for elderly patients is used to provide feedback and confirmation through the task compliance score, thereby obtaining the stage adaptation task.

[0012] In this embodiment, the nursing terminal is a mobile nursing terminal, which includes: a physiological parameter acquisition device, a task interaction interface, a posture sensor, and a communication module.

[0013] In this embodiment, the EEG device is a portable EEG device.

[0014] Secondly, this application provides a remote intelligent rehabilitation nursing system for elderly patients, used to execute a remote intelligent rehabilitation nursing method for elderly patients, the rehabilitation nursing system comprising: The risk scoring module is used to collect vital sign parameters of elderly patients through nursing terminals and determine risk level indicators in combination with the clinical examination results of elderly patients; The EEG signal processing module is used to collect EEG signals from elderly patients using an EEG analyzer, and to perform artifact suppression and frequency domain feature enhancement processing on the EEG signals to obtain signal feature segments. The task generation module is used to encode the signal feature segments according to the risk level index, thereby obtaining a matching feature vector for the rehabilitation task of the elderly patient. The task matching feature vector is used to prioritize the rehabilitation task of the elderly patient and obtain an adaptive task. The task adjustment module is used to remotely control the process of elderly patients performing the adaptation task, obtain a control coefficient, and adjust the nursing terminal of the elderly patients based on the control coefficient to obtain the stage adaptation task.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The technical solution provided in this application enables adaptive matching of remote rehabilitation tasks for elderly patients based on multimodal feature coding. Firstly, vital signs parameters of elderly patients are collected via a nursing terminal, and risk level indicators are determined in conjunction with clinical examination results. This transforms the multidimensional physiological state and potential health risks of elderly patients into quantifiable, tiered information. Secondly, electroencephalography (EEG) signals of elderly patients are collected using an EEG analyzer, and artifact suppression and frequency domain feature enhancement processing are performed on these signals to obtain signal feature fragments. This effectively improves the signal-to-noise ratio and discriminative power of EEG features in the time-frequency dimension, facilitating accurate characterization of the neural activity patterns and cognitive function states of elderly patients, thereby providing reliable neurophysiological feature support for rehabilitation task matching. Then, the signal feature fragments are processed according to the risk level indicators. The system performs feature encoding on segments to obtain matching feature vectors for rehabilitation tasks of elderly patients. These matching feature vectors are then used to prioritize candidate tasks in a pre-defined task library to obtain adaptive tasks. This enables adaptive mapping of multi-source heterogeneous health data of elderly patients to the rehabilitation task space, ensuring that the selection of rehabilitation tasks is accurately aligned with the patient's current overall condition, thus improving the targeting and rationality of task matching. Finally, the process of elderly patients performing adaptive tasks is remotely controlled to obtain control coefficients. Based on these control coefficients, the nursing terminals of elderly patients are adjusted to obtain stage-specific adaptive tasks. This allows for dynamic correction of rehabilitation task parameters as the patient's real-time condition changes, helping to avoid safety risks caused by task intensity mismatch and improving the safety and compliance of rehabilitation training.

[0016] In summary, the technical solution adopted in this application can achieve adaptive matching of remote rehabilitation tasks for elderly patients based on multimodal feature encoding, thereby improving the individualized accuracy of rehabilitation care. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a remote intelligent rehabilitation nursing method for elderly patients provided in this application; Figure 2This is an exemplary flowchart for determining matching feature vectors provided in this application; Figure 3 This is an exemplary flowchart for determining the control coefficient provided in this application; Figure 4 This is a schematic diagram illustrating the application scenario of the rehabilitation and nursing system provided in this application; Figure 5 This is a modular structure diagram of a remote intelligent rehabilitation and nursing system for elderly patients provided in this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Example 1: To better understand the technical solution of this application, the above technical solution will be described in detail below with reference to the accompanying drawings and specific embodiments. Figure 1 As shown in the figure, this is a flowchart of a remote intelligent rehabilitation nursing method for elderly patients according to this embodiment of the present application. The rehabilitation nursing method includes the following steps: In step S1, the vital signs parameters of elderly patients are collected through the nursing terminal, and the risk level indicators are determined in combination with the clinical examination results of the elderly patients.

[0021] It should be noted that, in this embodiment, the nursing terminal is a mobile nursing terminal, which includes: a physiological parameter acquisition device, a task interaction interface, a posture sensor, and a communication module; wherein, the physiological parameter acquisition device includes a photoplethysmography (PPG) sensor, a blood pressure sensor module, and a body temperature sensor module integrated into the housing of the mobile nursing terminal, used to collect vital signs parameters of elderly patients such as heart rate, blood oxygen saturation, systolic blood pressure, diastolic blood pressure, and body surface temperature; the task interaction interface is a touch screen display, used to present the content of rehabilitation tasks, action demonstration animations, and operation instructions to elderly patients. It receives touch operation feedback from elderly patients during rehabilitation tasks; the posture sensor is a six-axis inertial measurement unit integrated in the housing of the mobile nursing terminal, including a three-axis accelerometer and a three-axis gyroscope, used to collect limb motion acceleration, angular velocity and posture angle data of elderly patients during rehabilitation tasks; the communication module is a wireless communication module that supports Wi-Fi and cellular networks, used to upload the collected vital sign parameters, posture sensor data and operation records of the task interaction interface to the rehabilitation nursing system in real time, and to receive adaptation tasks and stage adaptation tasks issued by the rehabilitation nursing system.

[0022] In this embodiment, the risk level indicators for elderly patients are determined by collecting vital sign parameters through a nursing terminal and combining them with the patients' clinical examination results. Specifically, these indicators include: A multi-parameter feature matrix was constructed based on the vital signs parameters and clinical examination results of elderly patients; The multi-parameter feature matrix is ​​preprocessed, and then a risk prediction curve for elderly patients is generated based on the preprocessed multi-parameter feature matrix. The risk level index is determined by the risk prediction curve and the preset risk threshold.

[0023] It should be noted that the multi-parameter feature matrix in this application is a numerical two-dimensional table formed by organizing the vital signs parameters and clinical examination results of elderly patients according to a row and column structure. Each row of this multi-parameter feature matrix corresponds to one elderly patient, and each column corresponds to a parameter type, used to uniformly represent the multi-dimensional health status data of elderly patients. The risk prediction curve is a continuous curve with time as the independent variable and the risk prediction value as the dependent variable, used to describe the dynamic trend of the risk level of elderly patients over a future time period. The risk level index is an indicator value that combines the vital signs parameters and clinical examination results of elderly patients, used to represent the risk level of elderly patients at different time points, and can guide the generation and adaptation of rehabilitation tasks.

[0024] In practice, firstly, the physiological parameter acquisition device of the mobile nursing terminal acquires the heart rate, respiratory rate, blood pressure, and blood oxygen saturation values ​​of elderly patients. At the same time, the clinical examination results of elderly patients are retrieved through the hospital information system. The clinical examination results include comorbidity index, frailty score, cognitive scale score, and past medical history records. The acquired heart rate, respiratory rate, blood pressure, blood oxygen saturation, comorbidity index, frailty score, cognitive scale score, and past medical history records are organized into a row-column grid structure. Specifically, each vital sign parameter or clinical examination result is used as a column of the table, and all data of each elderly patient is used as a row of the table. The resulting row-column grid is then used as a multi-parameter feature matrix. Then, missing values ​​in the multi-parameter feature matrix can be filled using the mean imputation method, and the dimensions of the multi-parameter feature matrix can be unified by min-max normalization. Then, the redundant dimensions of the multi-parameter feature matrix can be reduced by the principal component analysis algorithm to obtain the preprocessed feature matrix. The preprocessed feature matrix can be input into the risk prediction model to output the risk prediction curve that changes over time. The risk prediction model refers to the support vector machine model, which is used to perform multi-dimensional correlation analysis on the feature matrix and can be trained by multi-source medical data.

[0025] Finally, the predicted risk value at each time point on the risk prediction curve is compared point-by-point with the preset risk thresholds. These preset risk thresholds include high-risk, medium-risk, and low-risk thresholds. The high-risk threshold defines a high-risk level, the medium-risk threshold defines a medium-risk level, and the low-risk threshold defines a low-risk level. For example, based on historical case statistics, the high-risk threshold can be set to 0.8, the medium-risk threshold to 0.5, and the low-risk threshold to 0.2. When the predicted risk value at a given time point is greater than 0.8, the risk level at that time point is determined to be high-risk; when it is greater than 0.5 and less than 0.8, the risk level at that time point is determined to be medium-risk; and when the predicted risk value at a given time point is greater than 0.2 and less than 0.5, the risk level at that time point is determined to be low-risk. This process iterates through all time points on the risk prediction curve, arranging the determined risk levels in chronological order, and then using the resulting risk level sequence as the risk level indicator.

[0026] In step S2, an electroencephalogram (EEG) device is used to collect the EEG signals of elderly patients, and the EEG signals are processed by artifact suppression and frequency domain feature enhancement to obtain signal feature segments.

[0027] It should be noted that, in this embodiment, the EEG device is a portable EEG device; a portable EEG device refers to a wearable EEG signal acquisition device worn on the head of an elderly patient. The portable EEG device includes multiple dry electrode sensors, a signal amplification circuit, an analog-to-digital conversion module, and a wireless data transmission module. Among them, the multiple dry electrode sensors are used to contact the scalp surface of the elderly patient to sense the weak potential changes generated by the electrical activity of brain neurons. The signal amplification circuit is used to differentially amplify the weak potential signals acquired by the dry electrode sensors to increase the signal amplitude. The analog-to-digital conversion module is used to convert the amplified analog potential signals into digital EEG signals. The wireless data transmission module is used to transmit the digital EEG signals to the rehabilitation nursing system or mobile nursing terminal through wireless communication.

[0028] In practice, an electroencephalogram (EEG) device is used to collect the EEG signals of elderly patients. Specifically, multiple dry electrode sensors of the portable EEG device are positioned on the corresponding locations of the elderly patient's scalp according to an internationally standardized lead system, ensuring stable contact between the dry electrode sensors and the scalp surface. After the portable EEG device is activated, the dry electrode sensors sense potential changes on the scalp surface at a set sampling frequency and transmit the sensed potential changes to a signal amplification circuit. The signal amplification circuit performs differential amplification on the received potential changes to obtain an amplified potential signal. The amplified potential signal is then transmitted to an analog-to-digital converter (ADC). The ADC performs discrete sampling and quantization encoding on the amplified potential signal at a set sampling frequency, converting the continuous analog potential signal into a discrete digital signal, thus obtaining the EEG signal. The EEG signal is a time-series signal with time as the horizontal axis and potential amplitude as the vertical axis, containing information on the electrical activity of brain neurons as well as various interference noise components.

[0029] In this embodiment, the EEG signal undergoes artifact suppression and frequency domain feature enhancement processing to obtain signal feature segments, specifically including: The EEG signal was subjected to baseline drift correction and decomposition reconstruction to obtain the corrected and reconstructed signal; The corrected and reconstructed signal is converted to the frequency domain, and then the spectral distribution characteristics in the frequency domain are subjected to low-frequency suppression processing to obtain the frequency domain enhanced signal. The frequency domain enhancement signal is segmented using a time-series segmentation algorithm to obtain signal feature segments.

[0030] It should be noted that the signal feature segment in this application refers to a set of feature vectors composed of multiple time segments, used for mapping and state discrimination in rehabilitation tasks. Baseline drift correction removes low-frequency drift and DC components from EEG signals through filtering and detrending techniques. Frequency domain enhancement signal refers to the EEG activity-related features obtained after processing key frequency bands in the spectral domain.

[0031] In practice, firstly, baseline drift correction is performed on the EEG signal using high-pass filtering. Then, wavelet decomposition is used to decompose and reconstruct the baseline-drift-corrected EEG signal, breaking it down into sub-signals of different frequency bands. Smoothing filtering is then used to zero out and suppress low-frequency sub-signals, yielding the corrected and reconstructed signal. Next, the corrected and reconstructed signal is transformed to the frequency domain using the Fast Fourier Transform algorithm to obtain its spectral distribution characteristics. Low-frequency noise components in these characteristics are suppressed to obtain the frequency-enhanced signal. Finally, a time-series segmentation algorithm is used to divide the frequency-enhanced signal into multiple time segments. This algorithm can be implemented using a sliding window, allowing for the extraction and standardization of time-frequency features in each time segment. The set of time-frequency features from all time segments is then used as the signal feature segment.

[0032] In step S3, the signal feature segments are feature-encoded according to the risk level index to obtain the matching feature vector of the rehabilitation task for elderly patients. The rehabilitation task for elderly patients is prioritized and the adaptive task is obtained by using the task matching feature vector.

[0033] Preferred, Reference Figure 2 As shown, this figure is an exemplary flowchart of determining the matching feature vector according to the present application. In this embodiment, the matching feature vector for the rehabilitation task of elderly patients is obtained by feature encoding the signal feature segment according to the risk level index. This can be achieved by the following steps: In step S31, the risk level index and the signal feature segment are time-stamped and normalized to obtain a fusion input matrix; In step S32, the fused input matrix is ​​subjected to scale convolutional encoding and weight combination to obtain a rehabilitation adaptation weight map. In step S33, the rehabilitation adaptation weight map is mapped and decoded to obtain the matching feature vector of the rehabilitation task for elderly patients.

[0034] It should be noted that the matching feature vector of the rehabilitation task for elderly patients in this application is a numerical vector that quantifies the matching strength between the current physiological state of the elderly patient and the candidate rehabilitation tasks in the preset task library. The rehabilitation adaptation weight map provides a clear matching weight distribution during the task library matching stage. Mapping decoding refers to the process of mapping a high-dimensional feature map into a low-dimensional interpretable index. In this embodiment, the fusion input matrix refers to the multi-source feature matrix after time alignment, interpolation padding, and dimension normalization. Scale convolutional coding refers to the coding strategy of using different one-dimensional convolution kernels in parallel to extract short-term and long-term dependent features.

[0035] In specific implementation, firstly, the risk level indicators and signal feature segments are time-stamped and normalized. This involves aligning timestamps from different acquisition points using a network time protocol, resampling low-sampling-rate data to a unified time grid using linear interpolation based on the alignment results, and eliminating dimensional differences by using min-max normalization for numerical features. Then, the interpolated and normalized feature vectors, arranged in chronological order, are row-ordered to obtain the fused input matrix. Next, a rehabilitation task adaptive generative network is used to perform scale convolutional encoding and weight combination on the fused input matrix. Specifically, batch normalization and non-linear activation are performed on each convolutional output through a one-dimensional convolutional kernel in the encoding channels of the rehabilitation task adaptive generative network. The convolutional features at each scale are then concatenated through channels to obtain an initial feature map. Subsequently, softmax is used to generate channel weights, which are then multiplied channel-by-channel with the initial feature map to achieve weight combination, resulting in a multi-channel rehabilitation adaptation weight map. Finally, the rehabilitation adaptation weight map is mapped and decoded in the decoding layer of the network. That is, the weight map is flattened and input into the fully connected layer for dimensionality reduction and nonlinear mapping. The output is converted into a set of numerical indicators by normalized activation function. The output vector is post-processed by mapping according to normalized range to obtain a quantized vector that can be directly used for task matching. Then, the quantized vector is used as the matching feature vector for rehabilitation tasks of elderly patients.

[0036] It should be noted that in this application, the adaptive generation network for rehabilitation tasks is a deep neural network based on a multimodal feature fusion algorithm and an adaptive attention mechanism. The multimodal feature fusion algorithm is used to map data from different sources, such as risk level indicators, EEG signal features, and rehabilitation execution records of elderly patients, to a unified feature space to ensure the comparability and complementarity between data. The adaptive attention mechanism is used to automatically identify key features that have a greater impact on the generation of rehabilitation tasks during the feature mapping process and assign them higher weights, thereby improving the relevance and individualization of the task generation results. In this embodiment, the adaptive generative network for rehabilitation tasks can be constructed by combining convolutional neural networks, recurrent neural networks, and fully connected networks with the vital signs parameters, clinical examination results, electroencephalogram (EEG) signals, and rehabilitation execution records of elderly patients as multi-source input data. Specifically, the convolutional neural network is used as a temporal feature encoder for EEG signals to extract local time-series features from the EEG data; the recurrent neural network is used as a temporal encoder for vital signs and clinical indicators to capture dependencies over long time spans; and the fully connected network is used as a structured data encoder for rehabilitation execution records to generate numerical representations aligned with other modalities. Finally, the adaptive generative network for rehabilitation tasks is obtained by training the multimodal features through an adaptive attention mechanism.

[0037] In this embodiment, the rehabilitation tasks for elderly patients are prioritized using the task matching feature vector, and the resulting adaptive tasks specifically include: The matching feature vector is matched with all rehabilitation tasks in the preset task library to obtain a fit sequence. The fitness sequence is subjected to risk weighting to obtain a weighted priority sequence; A task selection discrimination matrix is ​​constructed based on the weighted priority sequence, and then the selection probability distribution of candidate tasks is obtained through the task selection discrimination matrix. The adaptation task for elderly patients is output based on the selected probability distribution.

[0038] It should be noted that the adaptation task in this embodiment refers to a task adapted to the current rehabilitation stage of the elderly patient. The fit sequence is a numerical sequence of similarity matching strengths arranged by task, used to initially rank candidate tasks under safety constraints. Risk weighting is a process of converting risk level, resource constraints, and compliance history into weights that can be combined with similarity and then mathematically synthesizing them. The weighted priority sequence is a sequence of task priority scores that integrates similarity matching and multiple constraint weighting. The task selection discrimination matrix is ​​a two-dimensional scoring structure with tasks as rows and evaluation criteria as columns, used for multi-criteria normalization and column weighting. The selection probability distribution is used to control the participation ratio and safety boundary of different types of tasks. In addition, the preset task library refers to a collection of rehabilitation tasks stored in the rehabilitation nursing system. This library contains multiple rehabilitation tasks, each associated with corresponding task attribute information. These attributes include the training method, target body part, intensity level, duration, feedback method, and required equipment. Specifically, the training method represents the type of movement used in the task; the target body part represents the limb or functional area primarily targeted; the intensity level represents the training load; the duration represents the length of a single execution; the feedback method represents the method of providing prompts or assessments to the patient during the task; and the required equipment represents the instruments or assistive tools needed to perform the task. During the rehabilitation task generation process, each task in the preset task library participates in the similarity calculation of the matching feature vectors as a task attribute vector, thereby selecting the most suitable adaptive task from the candidate task set that best matches the elderly patient's current overall health status. The rehabilitation tasks in the preset task library can be pre-entered and configured by rehabilitation physicians or therapists according to clinical rehabilitation guidelines, and can also be dynamically added, deleted, or adjusted according to actual application scenarios.

[0039] In specific implementation, firstly, one-hot encoding is used to convert the task attributes (training method, target body part, intensity level, duration, feedback form, required equipment, etc.) of each rehabilitation task in the preset task library into numerical vectors. Then, min-max normalization is used to unify the dimensions of the numerical vectors. Next, the matching feature vectors of the elderly patient rehabilitation tasks are aligned according to the same feature dimension of the numerical vectors. Cosine similarity is used to calculate the similarity matching strength between the numerical vectors of each task and the matching feature vectors of the elderly patient rehabilitation tasks. Finally, the sequence of all similarity matching strengths is used as the fitness sequence. Secondly, interval mapping is used to map the risk level index of the elderly patients into risk weight coefficients, and then element-wise multiplication is used... The method involves multiplying the fitness sequence and risk weight coefficients item by item to obtain a weighted priority sequence. Then, a task selection discrimination matrix is ​​constructed based on the weighted priority sequence, i.e., with tasks as rows and evaluation criteria as columns (evaluation criteria include similarity matching strength, risk weight, resource load, fatigue budget, historical performance, etc.). Min-max normalization is then used to compress the columns to a uniform dimension, and the column vectors are weighted using the weighted priority sequence to obtain a comprehensive score vector for each task. Then, an exponential normalization function is used to transform the score into a selection probability distribution. Finally, a hierarchical task list is generated from the selection probability distribution using a combination of probability sampling and maximum value retention, and this hierarchical task list is used as the adaptation task.

[0040] In step S4, the process of the elderly patient performing the adaptation task is remotely controlled to obtain a control coefficient. Based on the control coefficient, the nursing terminal of the elderly patient is adjusted to obtain the stage adaptation task.

[0041] Preferred, Reference Figure 3 As shown, this diagram is an exemplary flowchart for determining the control coefficient according to the present application. In this embodiment, the process of remotely controlling the elderly patient's performance of the adaptation task based on the nursing terminal to obtain the control coefficient can be achieved through the following steps: In step S41, the task performance vector and EEG signal feature vector of the elderly patient when performing the adaptation task are determined; In step S42, modality fusion is performed on the task performance vector and the EEG signal feature vector based on a deep discriminant network to obtain a fused representation vector; In step S43, the fused representation vector is subjected to hierarchical risk discrimination to obtain the control coefficient.

[0042] It should be noted that the task performance vector in this embodiment is a multi-dimensional numerical sequence formed by combining limb movement data collected by posture sensors and operation feedback data recorded by the task interaction interface during the elderly patient's performance of the adaptation task. This task performance vector is used to quantitatively characterize the quality of the elderly patient's action completion and operational performance level during the rehabilitation task. The EEG signal feature vector is a set of multi-band power spectral density values ​​extracted from signal feature segments corresponding to the execution period of the adaptation task. This EEG signal feature vector is used to quantitatively characterize the elderly patient's neurophysiological response and cognitive load state during the rehabilitation task. The fusion representation vector is a comprehensive feature vector obtained by concatenating and attention-weighting the task performance vector and the EEG signal feature vector after feature extraction through a fully connected layer. This fusion representation vector is used to uniformly characterize the elderly patient's task performance and neurophysiological response in the same feature space. The control coefficient is a comprehensive score value obtained by performing hierarchical risk discrimination and scoring mapping on the fusion representation vector. This control coefficient is used to comprehensively characterize the elderly patient's physiological and cognitive control ability during the execution of the rehabilitation task and can provide a quantitative basis for real-time correction of subsequent rehabilitation task parameters.

[0043] In specific implementation, firstly, during the process of elderly patients performing adaptation tasks through the task interaction interface of the mobile nursing terminal, the posture sensor of the mobile nursing terminal collects the triaxial acceleration, triaxial angular velocity, and posture angle values ​​of the elderly patient's limbs in real time. At the same time, the task interaction interface records the time taken for the elderly patient to complete each task action and the operation feedback data. The collected triaxial acceleration, triaxial angular velocity, posture angle, action time, and operation feedback data are aligned according to timestamps, and all aligned values ​​are combined into a multidimensional numerical sequence, which is used as the task performance vector. Simultaneously, signal feature segments corresponding to the current adaptation task execution period are extracted from the signal feature segments output by the EEG signal processing module. The signal feature segments contain power spectral density values ​​of multiple frequency bands, and the extracted signal feature segments are used as EEG signal feature vectors.

[0044] Then, the task performance vector and EEG signal feature vector are input into a pre-trained deep discriminant network, which includes a first fully connected layer, a second fully connected layer, a third fully connected layer, and an attention fusion layer. The task performance vector is input into the first fully connected layer, which performs a linear transformation on it by multiplying the task performance vector by the weight matrix of the first fully connected layer and then adding it to the bias vector of the first fully connected layer to obtain a first transformed vector. This first transformed vector is then input into a non-linear activation function, which is a linear rectified function, to obtain a first feature vector. The EEG signal feature vector is input into the second fully connected layer, which performs a linear transformation on it by multiplying the EEG signal feature vector by the weight matrix of the second fully connected layer. The first feature vector is multiplied and then added to the bias vector of the second fully connected layer to obtain the second transformation vector. The second transformation vector is then input into a nonlinear activation function to obtain the second feature vector. The first and second feature vectors are then concatenated according to their feature dimensions, that is, the elements of the second feature vector are sequentially arranged at the end of the first feature vector to form a merged feature vector. The merged feature vector is then input into the attention fusion layer, which includes a third fully connected layer and an element-wise multiplication module. The third fully connected layer performs a linear transformation on the merged feature vector and then generates an attention weight vector through a flexible maximum function. Each element in the merged feature vector is multiplied by the corresponding weight coefficient in the attention weight vector to obtain a weighted fused feature vector. The weighted fused feature vector is then used as the fused representation vector.

[0045] Finally, the fused representation vector is input to the hierarchical risk discrimination module, which includes a fourth fully connected layer, a fifth fully connected layer, and a scoring mapping layer. The fused representation vector is input to the fourth fully connected layer for linear transformation and nonlinear activation to obtain the first risk feature vector. The first risk feature vector is then input to the fifth fully connected layer for linear transformation and nonlinear activation to obtain the second risk feature vector. The second risk feature vector is then input to multiple scoring prediction branches, including a fatigue scoring prediction branch, a task stability scoring prediction branch, and a risk propensity scoring prediction branch. Each scoring prediction branch consists of a fully connected layer. The system consists of three branches: a fatigue score prediction branch outputs a fatigue score value, a task stability score prediction branch outputs a task stability score value, and a risk tendency score prediction branch outputs a risk tendency score value. The fatigue score value, task stability score value, and risk tendency score value are weighted and summed according to preset weight coefficients to obtain the control coefficient. The weight coefficients are values ​​between zero and one, and the sum of the three weight coefficients is equal to 1. Optionally, in this embodiment, the initial value of the weight coefficients of the fatigue score value, task stability score value, and risk tendency score value can all be set to 1 / 3, which is not limited here.

[0046] In this embodiment, the nursing terminal for elderly patients is adjusted based on the control coefficient to obtain the stage adaptation task, which specifically includes: The task intensity, duration, and feedback frequency of elderly patients performing the adaptive task are adjusted in real time to obtain a task adjustment curve. Based on the changing trends of the control coefficient and the task correction curve, a fusion modeling analysis is performed to obtain a task compliance score. The nursing terminal for elderly patients is used to provide feedback and confirmation through the task compliance score, thereby obtaining the stage adaptation task.

[0047] It should be noted that the task correction curve in this embodiment is a continuous curve plotted with the task execution time point as the horizontal axis and the combined value of the changes in task intensity, duration, and feedback frequency as the vertical axis. This task correction curve is used to characterize the real-time correction trend and magnitude of the key parameters of the current adaptive task relative to the original set values ​​under the drive of the control coefficient. The task compliance score is a single quantitative score obtained by fusing the stability index of the task correction curve with the changing trend of the control coefficient through modeling and analysis. This task compliance score is used to comprehensively measure the overall degree of adaptation of elderly patients to the rehabilitation task plan in multiple dimensions such as physiological load, execution duration, and feedback rhythm. The stage adaptation task is a task plan formed by correcting or adjusting the task intensity level, duration, and feedback frequency values ​​of the current adaptive task. This stage adaptation task is used to replace the current adaptive task in the next execution cycle, so that the execution parameters of the rehabilitation task remain dynamically matched with the real-time status of the elderly patient, and the task parameters are progressively optimized with the rehabilitation process.

[0048] In practice, firstly, the task intensity level, duration, and feedback frequency values ​​are read from the task attributes of the current adaptive task. The task intensity level represents the training load intensity of the rehabilitation task, the duration represents the length of a single execution of the rehabilitation task, and the feedback frequency represents the time interval between providing operation prompts and effect feedback to the elderly patient during the execution of the rehabilitation task. Then, the control coefficient is applied to each of the task intensity level, duration, and feedback frequency values ​​individually: the task intensity level is multiplied by the control coefficient to obtain the corrected task intensity level, and the duration is multiplied by the control coefficient to obtain the corrected duration. The feedback frequency value is multiplied by the control coefficient to obtain the corrected feedback frequency value. The corresponding corrected values ​​are subtracted from the original task intensity level value, duration value, and feedback frequency value to obtain the changes in task intensity, duration, and feedback frequency. Three change curves are plotted on a coordinate system with the task execution time point as the horizontal axis and the changes in task intensity, duration, and feedback frequency as the vertical axes. The three change curves are then smoothed by taking the arithmetic mean of a preset number of data points before and after each data point on the curve as the new value for that point, thus removing local fluctuations in the curve, resulting in the task correction curve.

[0049] Then, a sliding window statistical algorithm is used to analyze the stability of the task correction curve. Specifically, a fixed-duration time window is set at the beginning of the task correction curve. The arithmetic mean and variance of all data points within this time window are calculated, and the arithmetic mean and variance are used as stability indicators for this window. The time window is then slid forward along the time axis by a fixed step, and the arithmetic mean and variance of the next time window are calculated. This sliding and calculation operation is repeated until the end of the time window reaches the end position of the task correction curve, resulting in a sequence of arithmetic mean and variance values ​​corresponding to multiple time windows. The arithmetic mean and variance values ​​are then exponentially smoothed to obtain a smoothed mean trend sequence and a smoothed variance trend sequence. The time series sequence of the control coefficient as a function of time is obtained, and the trend of the control coefficient time series is fused and analyzed with the trend of the task correction curve. Specifically, the value of each time point on the task correction curve is multiplied by the value of the corresponding time point in the control coefficient time series to obtain a fused value. All fused values ​​are arranged in chronological order to obtain a fused value sequence. Calculate the arithmetic mean and standard deviation of the fused numerical sequence. Subtract the standard deviation from the arithmetic mean as the first threshold, and add the standard deviation to the arithmetic mean as the second threshold. For each fused numerical value in the fused numerical sequence, determine the proportion of data points whose fused values ​​fall between the first and second thresholds out of the total number of data points. Use this proportion as the task compliance score.

[0050] Finally, the task compliance score is compared with a preset compliance threshold. The compliance threshold can be preset according to clinical rehabilitation assessment standards; for example, it can be set between 70% and 90% to define the elderly patient's adaptation to the rehabilitation task plan. If the task compliance score is greater than or equal to the compliance threshold, the elderly patient is considered to have adapted well to the current adaptive task, and the current adaptive task is directly used as the next stage of adaptation. If the task compliance score is less than the compliance threshold, the elderly patient is considered to have insufficient adaptation to the current adaptive task, and the current adaptive task needs to be adjusted. The adjustment method is based on the task compliance score... The difference between the score and the compliance threshold determines the adjustment range. This difference can be mapped to the adjustment ratio of the task intensity level value, duration value, and feedback frequency value. The task intensity level value is reduced, the duration value is shortened, and the feedback frequency value is increased using the adjustment ratio to obtain the adjusted task intensity level value, adjusted duration value, and adjusted feedback frequency value. The adjusted task intensity level value, adjusted duration value, and adjusted feedback frequency value are then recombined into a task plan. This task plan is used as a phase adaptation task and is sent to the task interaction interface of the mobile nursing terminal.

[0051] It should also be noted that the rehabilitation nursing system in this application is a remote rehabilitation nursing platform based on EEG and nursing terminals for multi-source data acquisition and intelligent rehabilitation task scheduling for elderly patients. It can achieve complete closed-loop management of rehabilitation nursing for elderly patients, from risk assessment and task generation to execution control and feedback adjustment. Preferably, [the system] can be referred to... Figure 4 As shown in the figure, this is a schematic diagram of the application scenario of the rehabilitation and nursing system provided in this application; wherein, the nursing terminal can be deployed on the elderly patient side to collect vital sign parameters and posture sensing data and receive the rehabilitation tasks issued; the electroencephalogram (EEG) device can be worn on the elderly patient's head to collect brain signals and transmit them to the rehabilitation and nursing system wirelessly; the mobile nursing terminal interacts with the rehabilitation and nursing system bidirectionally through a wireless network to realize the real-time issuance of rehabilitation tasks and the real-time feedback of execution data.

[0052] In summary, the technical solution adopted in this application can achieve adaptive matching of remote rehabilitation tasks for elderly patients based on multimodal feature encoding, thereby improving the individualized accuracy of rehabilitation care.

[0053] Example 2: This application provides a remote intelligent rehabilitation and nursing system for elderly patients, referencing... Figure 5 As shown, this figure is a modular structure diagram of a remote intelligent rehabilitation and nursing system for elderly patients according to this application. The rehabilitation and nursing system includes: The risk scoring module 100 is used to collect vital sign parameters of elderly patients through the nursing terminal and determine the risk level index in combination with the clinical examination results of elderly patients. The EEG signal processing module 200 is used to collect EEG signals from elderly patients using an EEG instrument, and to perform artifact suppression and frequency domain feature enhancement processing on the EEG signals to obtain signal feature segments. The task generation module 300 is used to encode the signal feature segments according to the risk level index, thereby obtaining a matching feature vector for the rehabilitation task of the elderly patient, and to evaluate the priority of the rehabilitation task of the elderly patient through the task matching feature vector to obtain an adaptive task. The task adjustment module 400 is used to remotely control the process of elderly patients performing the adaptation task, obtain a control coefficient, and adjust the nursing terminal of the elderly patients based on the control coefficient to obtain the stage adaptation task.

[0054] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0055] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0056] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A remote intelligent rehabilitation nursing method for elderly patients, characterized in that, The rehabilitation nursing method includes the following steps: The vital signs parameters of elderly patients are collected through nursing terminals, and the risk level indicators are determined in combination with the clinical examination results of elderly patients. Electroencephalography (EEG) was used to collect EEG signals from elderly patients. The EEG signals were then processed with artifact suppression and frequency domain feature enhancement to obtain signal feature segments. The signal feature segments are feature-encoded according to the risk level index to obtain the matching feature vector of the rehabilitation task for elderly patients. The rehabilitation task of elderly patients is prioritized by the task matching feature vector to obtain the adaptive task. The process of elderly patients performing the adaptation task is remotely controlled to obtain a control coefficient. The nursing terminal of the elderly patients is then adjusted based on the control coefficient to obtain the stage adaptation task.

2. The remote intelligent rehabilitation nursing method for elderly patients as described in claim 1, characterized in that, The risk level indicators for elderly patients are determined by collecting vital signs parameters through nursing terminals and combining them with clinical examination results. These indicators include: A multi-parameter feature matrix was constructed based on the vital signs parameters and clinical examination results of elderly patients; The multi-parameter feature matrix is ​​preprocessed, and then a risk prediction curve for elderly patients is generated based on the preprocessed multi-parameter feature matrix. The risk level index is determined by the risk prediction curve and the preset risk threshold.

3. The remote intelligent rehabilitation nursing method for elderly patients as described in claim 1, characterized in that, The EEG signal is subjected to artifact suppression and frequency domain feature enhancement processing to obtain signal feature segments, specifically including: The EEG signal was subjected to baseline drift correction and decomposition reconstruction to obtain the corrected and reconstructed signal; The corrected and reconstructed signal is converted to the frequency domain, and then the spectral distribution characteristics in the frequency domain are subjected to low-frequency suppression processing to obtain the frequency domain enhanced signal. The frequency domain enhancement signal is segmented using a time-series segmentation algorithm to obtain signal feature segments.

4. The remote intelligent rehabilitation nursing method for elderly patients as described in claim 1, characterized in that, Based on the risk level index, feature encoding is performed on the signal feature segments to obtain the matching feature vector for the rehabilitation task of elderly patients. Specifically, this includes: The risk level index and the signal feature segment are time-stamped and normalized to obtain a fusion input matrix; The fused input matrix is ​​subjected to scale convolutional encoding and weight combination to obtain a rehabilitation adaptation weight map; The rehabilitation adaptation weight map is mapped and decoded to obtain the matching feature vector of the rehabilitation task for elderly patients.

5. A remote intelligent rehabilitation nursing method for elderly patients as described in claim 1, characterized in that, The rehabilitation tasks for elderly patients are prioritized using the task matching feature vector, resulting in adaptive tasks that specifically include: The matching feature vector is matched with all rehabilitation tasks in the preset task library to obtain a fit sequence. The fitness sequence is subjected to risk weighting to obtain a weighted priority sequence; A task selection discrimination matrix is ​​constructed based on the weighted priority sequence, and then the selection probability distribution of candidate tasks is obtained through the task selection discrimination matrix. The adaptation task for elderly patients is output based on the selected probability distribution.

6. A remote intelligent rehabilitation nursing method for elderly patients as described in claim 1, characterized in that, The process of elderly patients performing the aforementioned adaptive tasks is remotely controlled via a nursing terminal, and the control coefficients obtained specifically include: Determine the task performance vector and EEG signal feature vector of elderly patients when performing adaptive tasks; Based on a deep discriminant network, modality fusion is performed on the task performance vector and the EEG signal feature vector to obtain a fused representation vector; The fused representation vector is subjected to hierarchical risk discrimination to obtain the control coefficient.

7. A remote intelligent rehabilitation nursing method for elderly patients as described in claim 1, characterized in that, Based on the aforementioned control coefficient, feedback adjustments are made to the nursing terminal for elderly patients to obtain the specific stage adaptation tasks, which include: The task intensity, duration, and feedback frequency of elderly patients performing the adaptive task are adjusted in real time to obtain a task adjustment curve. Based on the changing trends of the control coefficient and the task correction curve, a fusion modeling analysis is performed to obtain a task compliance score. The nursing terminal for elderly patients is used to provide feedback and confirmation through the task compliance score, thereby obtaining the stage adaptation task.

8. A remote intelligent rehabilitation nursing method for elderly patients as described in claim 1, characterized in that, The nursing terminal is a mobile nursing terminal, which includes: a physiological parameter acquisition device, a task interaction interface, a posture sensor, and a communication module.

9. A remote intelligent rehabilitation nursing method for elderly patients as described in claim 1, characterized in that, The EEG device is a portable EEG device.

10. A remote intelligent rehabilitation and nursing system for elderly patients, characterized in that, For performing a remote intelligent rehabilitation nursing method for elderly patients as described in any one of claims 1 to 9, the rehabilitation nursing system comprises: The risk scoring module is used to collect vital sign parameters of elderly patients through nursing terminals and determine risk level indicators in combination with the clinical examination results of elderly patients; The EEG signal processing module is used to collect EEG signals from elderly patients using an EEG analyzer, and to perform artifact suppression and frequency domain feature enhancement processing on the EEG signals to obtain signal feature segments. The task generation module is used to encode the signal feature segments according to the risk level index, thereby obtaining a matching feature vector for the rehabilitation task of the elderly patient. The task matching feature vector is used to prioritize the rehabilitation task of the elderly patient and obtain an adaptive task. The task adjustment module is used to remotely control the process of elderly patients performing the adaptation task, obtain a control coefficient, and adjust the nursing terminal of the elderly patients based on the control coefficient to obtain the stage adaptation task.

Images