An electronic medical record-based care system

Abstract

The application relates to the technical field of medical care, and provides a nursing system based on electronic medical records, which comprises a heterogeneous data integration module, which obtains multi-source heterogeneous electronic medical record data from a hospital HIS system, a wearable monitoring device and a nursing operation terminal and outputs the data after cleaning and standardization; a feature correlation module, which receives the data after cleaning and standardization to generate a space-time feature correlation matrix and stores a feature topology graph; a dynamic decision module, which processes the feature correlation matrix based on a deep space-time fusion network to generate a nursing decision vector; a scheme execution monitoring module, which collects nursing operation data streams in real time, calculates an execution deviation parameter and feeds back the execution deviation parameter; an abnormal intervention module, which analyzes the execution deviation parameter to trigger the dynamic decision module to execute nursing scheme reconstruction; a multi-terminal cooperation module, which converts the nursing decision vector into a device instruction set and feeds back a device state; and a safe storage module, which encrypts and stores the feature topology graph and a decision knowledge graph and provides model parameter version management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical care, and more particularly, the present application relates to a nursing system based on electronic medical records. BACKGROUND

[0002] In the field of medical care, with the continuous development of information technology, electronic medical record systems have gradually become popular, providing convenience for the storage and management of medical data. However, the existing nursing systems are mostly limited to simple data recording and query functions, lacking the ability to deeply integrate and analyze multi-source heterogeneous data. In actual nursing processes, medical staff need to manually review a large amount of scattered data, making it difficult to quickly and accurately make nursing decisions. In addition, the execution of nursing operations lacks real-time monitoring and dynamic adjustment mechanisms, and once there is a deviation from the standard operation, it is difficult to discover and correct in a timely manner, which may affect the quality of nursing and patient safety. At the same time, the security and privacy of nursing data also face many challenges, and there is a risk of tampering or leakage during data storage and transmission.

[0003] In the process of implementing the embodiments of the present application, the present application finds that at least the following problems or defects exist in the prior art: the existing nursing system lacks the ability to integrate multi-source heterogeneous data, and cannot effectively mine the potential value of data to assist nursing decisions; it lacks the function of real-time monitoring of nursing operation execution and dynamically adjusting the nursing scheme according to the deviation, making it difficult to meet the complex and variable clinical nursing needs; the data security and privacy protection measures are not perfect, and the safety of patient information cannot be fully guaranteed. SUMMARY

[0004] The present application provides a nursing system based on electronic medical records, comprising:

[0005] a heterogeneous data integration module for obtaining multi-source heterogeneous electronic medical record data from a hospital HIS system, a wearable monitoring device and a nursing operation terminal, and outputting the data to a feature correlation module after cleaning and standardization;

[0006] a feature correlation module for receiving the data after cleaning and standardization and generating a spatio-temporal feature correlation matrix, outputting the matrix to a dynamic decision module, and storing a feature topology graph to a secure storage module;

[0007] a dynamic decision module for processing the spatio-temporal feature correlation matrix based on a deep spatio-temporal fusion network, generating a nursing decision vector, and distributing the vector to a scheme execution monitoring module and a multi-end collaboration module;

[0008] a scheme execution monitoring module for real-time acquisition of nursing operation data stream, calculation of execution deviation parameter and feedback to an abnormal intervention module and a dynamic decision module;

[0009] an abnormal intervention module for analyzing the execution deviation parameter, triggering the dynamic decision module to reconstruct the care plan when the deviation exceeds a threshold value;

[0010] a multi-terminal coordination module for converting the care decision vector into a device instruction set, arbitrating conflicting instructions, and distributing the instructions to terminal devices, and feeding back device status to the dynamic decision module;

[0011] a secure storage module for encrypting the feature topology graph and decision knowledge graph, and providing model parameter version management for the dynamic decision module.

[0012] Further, the heterogeneous data integration module performs data cleaning, including the following steps:

[0013] Based on the Markov chain model, missing data is estimated by the previous state transition probability, historical mean and Gaussian noise;

[0014] The complete data is segmented and nonlinearly normalized. When the data value is greater than or equal to a preset reference threshold, it is mapped to the [0, 1] interval. When the data value is less than the reference threshold, it is mapped to the [0, 0.5] interval.

[0015] Further, the feature correlation module performs the following steps:

[0016] The data after cleaning and standardization is constructed into a three-dimensional feature tensor according to time slices, ward areas and physiological indicators.

[0017] Based on the three-dimensional feature tensor, a spatio-temporal attention weight matrix is calculated, as shown in the following formula:

[0018] ,

[0019] wherein, represents the query vector of the i-th time slice, represents the key vector of the j-th spatial position, represents the query projection matrix, represents the key projection matrix, represents the scaling factor, and the three-dimensional feature tensor wherein 24 represents the number of 24-hour time slices, W represents the total number of ward areas, and C=12 represents the number of physiological indicator dimensions;

[0020] After the spatio-temporal attention weight matrix is spliced with the three-dimensional feature tensor, it is input into a three-dimensional convolutional network for feature extraction.

[0021] Further, the three-dimensional convolutional network performs dynamic kernel updating, including the following steps:

[0022] Real-time monitoring of the feature change amount of the three-dimensional feature tensor;

[0023] The convolution kernel weight is dynamically adjusted by a gating cycle unit, wherein the current weight is determined by the weight of the previous moment and a one-dimensional vector of feature changes.

[0024] Further, the deep spatio-temporal fusion network processing procedure comprises:

[0025] The physiological feature vector and the environmental parameter vector are extracted from the spatio-temporal feature correlation matrix respectively;

[0026] The physiological feature vector and the environmental parameter vector are spliced, and the cross-modal correlation weight is calculated through a learnable parameter matrix;

[0027] Based on the cross-modal correlation weight, the physiological feature vector is subjected to multi-scale convolution processing, and features of different granularities are extracted by using large, medium and small convolution kernel sizes respectively;

[0028] The multi-scale features are fused through a gating mechanism to generate a final nursing decision vector.

[0029] Further, the scheme execution monitoring module calculates an execution deviation degree parameter, which comprises:

[0030] The actual nursing operation is decomposed into an atomic event sequence and aligned with a preset standard event sequence;

[0031] The minimum path difference between the actual sequence and the standard sequence is calculated through a dynamic time warping algorithm;

[0032] Based on the minimum path difference and the sequence length of the actual sequence and the standard sequence, a window deviation degree is calculated through an exponential decay function, which is used to quantify the operation deviation.

[0033] Further, when the deviation degree exceeds a threshold value, the dynamic decision module is triggered to reconstruct the nursing scheme, which comprises:

[0034] A sliding window length and a time decay weight coefficient are set;

[0035] In the sliding window, the window deviation degree is accumulated based on the time decay weight to obtain a cumulative deviation degree;

[0036] When the cumulative deviation degree and the current window deviation degree both exceed a preset threshold value, the dynamic decision module is triggered to reconstruct the nursing scheme.

[0037] Further, the multi-terminal collaborative module converts the nursing decision vector into a device instruction set, which comprises the following steps:

[0038] An instruction feature vector is extracted from the nursing decision vector, and an instruction delay time is obtained;

[0039] The urgency component is calculated based on a delay time, the importance component is calculated based on an instruction feature vector and a preset severity coefficient, and the instruction priority is obtained by weighted summation;

[0040] If there are concurrent conflict instructions, the instruction with shorter execution time or higher priority is selected for execution according to the priority difference.

[0041] Further, the fusion of multi-scale features through the gating mechanism to generate the final nursing decision vector comprises:

[0042] The orthogonality loss between the multi-scale feature matrices is calculated, and the orthogonality loss is the square sum of the Frobenius norm of the transpose product of each scale feature matrix.

[0043] Further, the total loss function of the fusion of multi-scale features through the gating mechanism comprises a weighted sum of a cross-entropy loss and an orthogonality loss, and the weight of the orthogonality loss is 0.01.

[0044] The above embodiments of the present application have at least the following beneficial effects:

[0045] The nursing system of the present application can realize efficient integration and standardized processing of multi-source heterogeneous data, and generate accurate nursing decision vectors through deep spatio-temporal fusion network for deep analysis of data, so as to provide scientific and reasonable nursing suggestions for medical staff, improve the accuracy and timeliness of nursing decisions. At the same time, the system can collect real-time nursing operation data stream, calculate and feedback execution deviation parameters, trigger dynamic reconstruction of nursing scheme, ensure the standardization and consistency of nursing operation, reduce nursing risk and improve nursing quality.

[0046] In addition, the multi-terminal collaboration module in the system can convert the nursing decision vector into a device instruction set to realize intelligent collaborative work of nursing equipment and improve nursing efficiency. The secure storage module uses encryption technology to store the feature topology graph and the decision knowledge graph, and provides model parameter version management to ensure the security and integrity of data, which can provide strong support for stable operation of the system and reliable traceability of data. BRIEF DESCRIPTION OF DRAWINGS

[0047] The above and other objects, features and advantages of the exemplary embodiments of the present application will be more apparent from the following detailed description taken in conjunction with the accompanying drawings, in which:

[0048] Figure 1 The structure schematic diagram of the nursing system based on electronic medical record provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0049] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided solely to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. Rather, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0050] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software.

[0051] It should be noted that any number of elements in the drawings is for illustration only and not for limitation, and any naming is only for distinction and does not have any limiting meaning.

[0052] Reference below Figure 1 , Figure 1 This is a schematic diagram of the structure of a nursing system based on electronic medical records provided by an embodiment of the present invention. Figure 1 As shown, a nursing system based on electronic medical records includes: a heterogeneous data integration module (1), which is used to obtain multi-source heterogeneous electronic medical record data from a hospital HIS system, wearable monitoring equipment and a nursing operation terminal, and output the data to a feature association module after cleaning and standardization; a feature association module (2), which is used to receive the cleaned and standardized data and generate a spatiotemporal feature association matrix, output the matrix to a dynamic decision module, and store the feature topology map in a secure storage module; a dynamic decision module (3), which is used to process the spatiotemporal feature association matrix based on a deep spatiotemporal fusion network, generate a nursing decision vector, and distribute the vector to a program execution monitoring module. A control module and a multi-terminal collaboration module; a plan execution monitoring module (4) for real-time collection of nursing operation data streams, calculation of execution deviation parameters and feedback to the abnormal intervention module and the dynamic decision module; an abnormal intervention module (5) for analyzing the execution deviation parameters and triggering the dynamic decision module to reconstruct the nursing plan when the deviation exceeds a threshold; a multi-terminal collaboration module (6) for converting the nursing decision vector into a device instruction set, distributing it to the terminal device after arbitration of conflicting instructions, and feeding back the device status to the dynamic decision module; a secure storage module (7) for encrypting and storing the feature topology map and decision knowledge map, and providing model parameter version management for the dynamic decision module.

[0053] It should be noted that the nursing system of the present application realizes the processing of electronic medical record data and the generation of nursing decisions through multiple interactive modules. The heterogeneous data integration module is responsible for obtaining multi-source heterogeneous electronic medical record data from the HIS system (hospital information system, a system for managing the daily operation and patient information of the hospital), wearable monitoring devices (such as smart bracelets, smart mattresses, etc., for real-time monitoring of patient vital signs and other data), and nursing operation terminals (devices used by medical staff for recording nursing operation information, etc.), and outputs the data to the feature correlation module after standardization. The feature correlation module receives the data after cleaning and standardization, generates a spatiotemporal feature correlation matrix, which is used to describe the correlation of the data in time and space, and outputs it to the dynamic decision module, while storing the feature topology graph describing the relationship between the features to the secure storage module. The dynamic decision module processes the feature correlation matrix based on the deep spatiotemporal fusion network to generate a nursing decision vector, which is distributed to the scheme execution monitoring module and the multi-end collaborative module. The scheme execution monitoring module collects nursing operation data streams in real time, calculates the execution deviation parameter (used to measure the deviation between actual nursing operation and standard operation), and feeds back the parameter to the abnormal intervention module and the dynamic decision module. The abnormal intervention module analyzes the execution deviation parameter, and when it finds that the deviation is too large, it triggers the dynamic decision module to execute nursing scheme reconstruction to adjust the nursing scheme. The multi-end collaborative module converts the nursing decision vector into a device instruction set (a set of instructions for controlling nursing devices), arbitrates and distributes it to terminal devices, and feeds back the device status to the dynamic decision module. The secure storage module is responsible for encrypting the feature topology graph and the decision knowledge graph (a graph structure for storing decision-related knowledge), and providing model parameter version management for the dynamic decision module to ensure the security and traceability of the data.

[0054] Specifically, the data cleaning performed by the heterogeneous data integration module includes the following steps:

[0055] Based on the Markov chain model, the missing data is estimated by the previous state transition probability, the historical mean value and the Gaussian noise;

[0056] The complete data is segmented and nonlinearly normalized, when the data value is greater than or equal to a preset reference threshold, it is mapped to the [0, 1] interval; when the data value is less than the reference threshold, it is mapped to the [0, 0.5] interval.

[0057] In the execution of data cleaning, the heterogeneous data integration module first estimates missing data through a Markov chain model. The Markov chain is a mathematical model used to describe the transition probabilities of system states. In this system, it estimates the missing data by calculating the transition probability of the current state from the previous state, thereby ensuring the integrity of the data. Then, a piecewise nonlinear function is used to normalize the data. Normalization is to scale the data proportionally so that it falls within a small range, such as 0 to 1. Here, the piecewise nonlinear function uses different function forms for different intervals of characteristic values. When the characteristic value is greater than or equal to a certain reference threshold, one function form is used; when the characteristic value is less than the reference threshold, another function form is used. This processing method can better adapt to the distribution characteristics of different characteristic values and improve the usability of the data. The feature correlation module sets the dimensions of the three-dimensional feature tensor to the number of 24-hour time slices, the total number of ward areas, and the number of physiological indicator dimensions. This means that the system divides a day into 24 time slices, considers all ward areas, and 12 physiological indicators to comprehensively describe the patient's state. In calculating the spatio-temporal attention weight matrix, the query vector, key vector, and corresponding projection matrix are used to calculate the weight through the scaled dot-product attention mechanism to highlight important spatio-temporal features. The dynamic decision module, in processing the feature correlation matrix, extracts physiological feature vectors and environmental parameter vectors through a deep spatio-temporal fusion network, calculates cross-modal correlation weights, performs multi-scale convolution processing, and performs gated fusion of multi-scale features to finally generate a nursing decision vector. These steps comprehensively consider physiological and environmental factors and improve the accuracy of decision-making through the fusion of multi-scale features.

[0058] Preferably, in the heterogeneous data integration module, for the step of estimating missing data through a Markov chain model, the calculation method of transition probability can be further optimized, such as introducing more historical data to improve the accuracy of estimation. In the feature correlation module, for the construction of the three-dimensional feature tensor, the total number of ward areas can be dynamically adjusted according to the number of patients in different ward areas to more flexibly adapt to the actual operation of the hospital. In the dynamic decision module, for the processing flow of the deep spatio-temporal fusion network, a feature verification step can be added before generating the nursing decision vector to ensure the accuracy and reliability of the extracted features. In addition, in the scheme execution monitoring module, when calculating the execution deviation degree, more evaluation indicators such as the time efficiency of nursing operations and the accuracy of operations can be introduced to more comprehensively evaluate the execution of nursing operations.

[0059] In some embodiments, the heterogeneous data integration module performs data cleaning, including:

[0060] Step one: Estimate missing data through a Markov chain model, the calculation formula is:

[0061] ,

[0062] where denotes the transition probability from the previous state to the current state , denotes the feature value at the previous time step, denotes the historical mean of the feature, denotes the random noise following a Gaussian distribution, denotes the noise variance.

[0063] Step 2: Normalize the data using piecewise nonlinear functions, when perform:

[0064] ,

[0065] when perform:

[0066] ,

[0067] where k = 2.5 denotes the steepness factor, denotes the feature reference threshold, and x denotes the original feature value.

[0068] It should be noted that the heterogeneous data integration module will first perform data cleaning when processing data. This is to remove noise and fill in missing values in the data, thereby improving the quality and usability of the data. The first step of data cleaning is to estimate missing data through a Markov chain model. Markov chain is a stochastic process model that assumes that the future state of a system only depends on the current state, and has nothing to do with the past state. In this system, the Markov chain model can be used to predict missing data points based on existing data sequences. Specifically, the model calculates the transition probability from the previous state to the current state, and combines the historical mean of the feature and the random noise following a Gaussian distribution to estimate the missing value. Gaussian distribution is a common probability distribution, and the random noise is used to simulate the uncertainty in the data. The second step is to normalize the data using a piecewise nonlinear function. Normalization is to convert the data to a specific range, so that the data of different features can be compared and processed on a unified scale. The piecewise nonlinear function is a function that processes different functions according to different intervals of input data, which can better adapt to the nonlinear characteristics of the data, making the data processing more flexible and accurate.

[0069] In particular, the calculation of transition probabilities is crucial when estimating missing data in a Markov chain model. Transition probabilities represent the likelihood of moving from one state to another. In the present system, by analyzing historical data sequences, a transition probability matrix between each state can be obtained. For example, if the current feature value is and the previous time's feature value is , the transition probability can be estimated by counting the frequency of appearing when appears. The historical mean of a feature is obtained by calculating the average value of the feature in historical data, which reflects the average level of the feature. Gaussian distributed random noise is used to simulate random fluctuations in the data, and the variance can be adjusted according to the degree of fluctuation of the data. In the piecewise nonlinear function, the steepness factor k and the feature reference threshold are two important parameters. The steepness factor k determines the steepness of the function, and a larger k value will make the function change more sharply near the threshold. The feature reference threshold is a value used to divide the data interval. When the original feature value x is greater than or equal to , a function form is used; when x is less than , another function form is used. This piecewise processing method can better adapt to the distribution characteristics of different feature values, improving the processing effect of the data.

[0070] Preferably, in the Markov chain model, more historical states can be introduced to estimate missing data, not just the previous state. For example, the previous two or more states can be considered to calculate the transition probability, thereby improving the accuracy of the estimate. For the piecewise nonlinear function, different function forms and parameters can be selected according to the specific application scenario and data characteristics. For example, the value of the steepness factor k can be adjusted to better suit the specific feature distribution. In addition, other types of noise distribution can be introduced to replace the Gaussian distribution to better simulate the uncertainty in the data. For example, if the data fluctuation presents a sharp peak and thick tail characteristic, other distributions such as Levy distribution can be considered to replace the Gaussian distribution.

[0071] Further, the feature correlation module performs the following steps:

[0072] The data after cleaning and standardization is constructed into a three-dimensional feature tensor according to time slice, ward area and physiological index dimensions;

[0073] Based on the three-dimensional feature tensor, a spatio-temporal attention weight matrix is calculated to quantify the importance of different spatio-temporal features;

[0074] After concatenating the spatio-temporal attention weight matrix with the three-dimensional feature tensor, input the three-dimensional convolutional network for feature extraction.

[0075] In some embodiments, the three-dimensional feature tensor is constructed where 24 represents the number of 24-hour time slices, W represents the total number of ward areas, and C=12 represents the number of physiological indicator dimensions.

[0076] The spatio-temporal attention weight matrix is calculated based on the tensor T:

[0077] ,

[0078] where, represents the query vector of the i-th time slice, represents the key vector of the j-th spatial location, represents the query projection matrix, represents the key projection matrix, represents the scaling factor;

[0079] After concatenating the weight matrix A with the tensor T, input the three-dimensional convolutional network, and the operation expression is:

[0080] ,

[0081] where represents the channel dimension concatenation operation, represents the convolution kernel size of the time, space, and feature dimensions, represents the sliding step of each dimension;

[0082] It should be noted that the main function of the feature correlation module is to convert the standardized data into a correlation matrix that can reflect the spatio-temporal features of the data, and store it for subsequent use. This process involves two key steps: constructing a three-dimensional feature tensor and calculating a spatio-temporal attention weight matrix. The three-dimensional feature tensor is a multi-dimensional data structure that can describe the distribution of data in three dimensions of time, space, and features. The spatio-temporal attention weight matrix is used to measure the importance and correlation between different time slices and spatial locations. In this way, those spatio-temporal features that are more critical for nursing decisions can be highlighted, providing more valuable information for subsequent decision-making.

[0083] Specifically, when constructing the three-dimensional feature tensor, its dimensions are set to the number of 24-hour time slices, the total number of ward areas, and the number of physiological indicator dimensions. This means that the system divides a day into 24 time slices, considers all ward areas, and 12 physiological indicators to comprehensively describe the patient's state. When calculating the spatio-temporal attention weight matrix, the query vector, key vector, and corresponding projection matrix are used to calculate the weight through the scaled dot-product attention mechanism to highlight important spatio-temporal features. The query vector and key vector are vectors extracted from the tensor, representing the features of the time slice and spatial position, respectively. The projection matrix is used to map these vectors into a suitable feature space for effective calculation. The scaling factor is used to prevent the problem of gradient vanishing caused by excessively large numerical values during calculation, and is usually the square root of the dimension of the query vector. In this way, a weight matrix can be obtained that reflects the degree of association between different time slices and spatial positions, thereby providing guidance for subsequent convolution operations.

[0084] Preferably, when constructing the three-dimensional feature tensor, the number of time slices, the division of ward areas, and the selection of physiological indicators can be flexibly adjusted according to the actual medical scene and data characteristics. For example, in some scenarios that require more fine time resolution, a day can be divided into more time slices, such as every half hour. For the division of ward areas, dynamic adjustment can be made according to the size of the ward, the number of patients, and the actual needs of nursing work. In the selection of physiological indicators, in addition to the 12 common physiological indicators, some special physiological indicators can be added according to the monitoring needs of specific diseases. When calculating the spatio-temporal attention weight matrix, different attention mechanisms can be tried, such as the multi-head attention mechanism, which can capture the association between spatio-temporal features from different angles, thereby improving the model's understanding ability of complex data. In addition, the parameters of the projection matrix can be optimized by learning during the training process to adjust its value to better adapt to the distribution characteristics of the data.

[0085] In some embodiments, the three-dimensional convolutional network performs dynamic kernel update including the following steps:

[0086] Monitoring the feature change amount of the three-dimensional feature tensor in real time;

[0087] Dynamically adjusting the convolution kernel weight through the gated recurrent unit, wherein the current weight is determined by the previous weight and the one-dimensional vector of the feature change amount.

[0088] Specifically, the three-dimensional convolutional network performs dynamic kernel update including:

[0089] Monitoring the feature change amount ,

[0090] where h = 24 represents the time dimension, w = W represents the spatial dimension, and c = 64 represents the number of feature channels;

[0091] The convolution kernel weight is adjusted by a gated recurrent unit:

[0092] ,

[0093] where, represents a gated recurrent network with 128 hidden units, represents the convolution kernel weight at the previous time step, represents the feature change amount, is flattened into a one-dimensional vector.

[0094] It should be noted that the three-dimensional convolution network plays a key role in the feature correlation module. It extracts more representative and discriminative features by performing convolution operations on the spatio-temporal feature tensor. The dynamic kernel update mechanism is an important feature of this network, which can dynamically adjust the weights of the convolution kernel according to the changes of the input features, thereby enhancing the adaptability of the network to different input patterns. The update of the convolution kernel weight is realized by a gated recurrent unit (GRU), which is a neural network structure commonly used to process sequence data. It can remember the previous state information and adjust the current weight accordingly. This dynamic updating mechanism enables the network to better capture the changes of spatio-temporal features, improving the accuracy and robustness of feature extraction.

[0095] Specifically, the dynamic kernel update mechanism first monitors the feature change amount, which is achieved by calculating the difference between the input features at different time steps. The dimensions of the feature change amount include the time dimension, the spatial dimension, and the number of feature channels, corresponding to the number of time slices, the number of ward areas, and the number of features, respectively. Then, the gated recurrent unit (GRU) is used to adjust the convolution kernel weight. The GRU receives the feature change amount at the current time step and the convolution kernel weight at the previous time step as input, and outputs the updated convolution kernel weight. The GRU contains multiple hidden units, and the number of these hidden units can be set according to actual needs, for example, 128 in this system. In this way, the convolution kernel weight can dynamically adapt to the changes of the input features, thereby improving the network's ability to extract spatio-temporal features. In addition, the update of the convolution kernel weight also involves the operation of flattening the feature change amount into a one-dimensional vector, because the input of the GRU needs to be a one-dimensional vector form, which can facilitate the input of the multi-dimensional feature change amount into the GRU for processing.

[0096] Preferably, more context information can be introduced to improve the accuracy of weight updating when monitoring the amount of feature change. For example, in addition to considering the feature change at the current time and the previous time, the feature change at the previous few times can also be combined to calculate the feature change amount, thereby providing more historical information for weight updating. For the number of hidden units of the gated recurrent unit (GRU), it can be adjusted according to the complexity of the data and the training effect of the network. If the spatio-temporal feature change of the data is relatively complex, the number of hidden units can be appropriately increased to enhance the memory and adaptability of the network. In addition, other types of recurrent neural network structures can also be tried to replace GRU, such as long short-term memory network (LSTM), which has better performance in processing long sequence data. In the updating process of the convolution kernel weight, a regularization mechanism such as weight decay can also be introduced to prevent overfitting phenomenon in the weight updating process, thereby improving the generalization ability of the network.

[0097] In some embodiments, the deep spatio-temporal fusion network processing procedure comprises:

[0098] extracting a physiological feature vector and an environmental parameter vector from the spatio-temporal feature correlation matrix, respectively;

[0099] concatenating the physiological feature vector and the environmental parameter vector, and calculating a cross-modal correlation weight through a learnable parameter matrix;

[0100] based on the cross-modal correlation weight, performing multi-scale convolution processing on the physiological feature vector, and extracting features of different granularities using large, medium and small convolution kernel sizes, respectively;

[0101] fusing multi-scale features through a gating mechanism to generate a final nursing decision vector.

[0102] Specifically, the physiological feature vector and the environmental parameter vector are extracted, where represents a 64-dimensional real number vector space;

[0103] the cross-modal correlation weight is calculated as:

[0104] ,

[0105] where represents a vector concatenation operation, represents a learnable parameter matrix, represents a sigmoid activation function;

[0106] multi-scale convolution processing is performed as:

[0107] ,

[0108] ,

[0109] ,

[0110] wherein, denotes the tensor product operation, kernel = 5 denotes a 5x5x5 convolution kernel; is a feature tensor extracted by a 5x5x5 convolution kernel, is a feature tensor extracted by a 3x3x3 convolution kernel, is a feature tensor extracted by a 1x1x1 convolution kernel, is a cross-modal correlation weight.

[0111] The gated fusion multi-scale feature is:

[0112] ,

[0113] ,

[0114] wherein, denotes a fusion weight matrix, denotes an element-wise multiplication operation, denotes a feature concatenation operation, is a gating signal, is an output feature tensor.

[0115] It is necessary to note that the deep spatio-temporal fusion network is the core module in the system for processing feature correlation matrix and generating nursing decision vector. It fuses physiological features and environmental parameters, extracts multi-scale features, and performs gated fusion through a series of complex processing steps, finally generating accurate nursing decision vector. This process involves multiple key steps, including extraction of physiological feature vector and environmental parameter vector, calculation of cross-modal correlation weight, multi-scale convolution processing, and gated fusion. The physiological feature vector and environmental parameter vector represent the physiological state of the patient and the parameters of the environment, respectively, and their fusion can provide more comprehensive information for nursing decision. Cross-modal correlation weight is used to measure the correlation degree between different modal data, while multi-scale convolution processing can capture feature information of different scales, and gated fusion further optimizes the combination of features, ensuring that the final generated nursing decision vector has high accuracy and practicality.

[0116] In particular, the deep spatio-temporal fusion network first extracts a physiological feature vector and an environmental parameter vector, both of which are 64-dimensional real-valued vector spaces. The physiological feature vector contains physiological-related data such as the patient's vital signs, medical history, etc., while the environmental parameter vector covers environmental factors such as temperature, humidity, and illumination in the ward. Next, when calculating the cross-modal correlation weight, the physiological feature vector and the environmental parameter vector are combined through vector concatenation and weighted by a learnable parameter matrix, and finally transformed by a sigmoid activation function for non-linear transformation to obtain the cross-modal correlation weight. This weight reflects the correlation between physiological features and environmental parameters, which helps subsequent feature fusion. In the multi-scale convolution processing stage, different sizes of convolution kernels (such as 5x5x5, 3x3x3, and 1x1x1) are used to perform convolution operations on the feature tensor to extract features of different scales. The size and number of these convolution kernels can be adjusted according to the characteristics of the actual data. Finally, in the gating fusion stage, a fusion weight matrix is used to weight and combine the multi-scale features, and element-wise multiplication is performed to obtain the final feature representation, which in turn generates a nursing decision vector.

[0117] Preferably, when extracting the physiological feature vector and the environmental parameter vector, different feature extraction methods can be selected according to the specific application scenario and data characteristics. For example, for the physiological feature vector, an autoencoder in deep learning can be introduced to extract more representative features. For the environmental parameter vector, sensor network data fusion technology can be combined to improve the accuracy and reliability of the environmental parameters. When calculating the cross-modal correlation weight, more complex neural network structures such as multi-layer perceptron (MLP) can be used instead of simple learnable parameter matrices to better capture the complex relationship between physiological features and environmental parameters.

[0118] Further, in the multi-scale convolution processing, attention mechanisms can be introduced to enable the network to automatically learn the importance of different scale features, further improving the effectiveness of feature extraction. In the gating fusion stage, different fusion strategies such as attention-based fusion or graph neural network-based fusion can be tried to better integrate multi-scale features and generate more accurate nursing decision vectors.

[0119] In some embodiments, the scheme execution monitoring module calculates an execution deviation parameter, including:

[0120] The actual nursing operation is decomposed into an atomic event sequence and aligned with a preset standard event sequence;

[0121] The minimum path difference between the actual sequence and the standard sequence is calculated by a dynamic time warping algorithm;

[0122] Based on the minimum path difference and the sequence length of the actual sequence and the standard sequence, the window deviation is calculated by an exponential decay function, which is used to quantify the operation deviation.

[0123] In some embodiments, the scheme execution monitoring module calculates the execution deviation, including:

[0124] Decomposing the nursing operation into atomic event sequences and standard sequences where m represents the actual sequence length, and n represents the standard sequence length.

[0125] Dynamic time warping calculates the minimum path difference:

[0126] ,

[0127] where represents the actual execution of the ith feature value, represents the standard jth feature value, represents the minimum path difference between the actual sequence and the standard sequence, i and j correspond to the positions in the actual sequence and the standard sequence respectively.

[0128] Calculate the window deviation:

[0129] ,

[0130] where 0.5 represents the sensitivity adjustment coefficient, represents the maximum value of the sequence length, is the window deviation.

[0131] It should be noted that the core function of the scheme execution monitoring module is to collect the nursing operation data stream in real time and calculate the execution deviation parameter, so as to quantitatively evaluate the execution of the nursing operation. This process involves decomposing the nursing operation into atomic event sequences and standard sequences, calculating the minimum path difference by dynamic time warping algorithm, and then obtaining the window deviation. Atomic event sequences refer to the events occurring in sequence in the actual nursing operation process, while standard sequences refer to the ideal event sequences pre-set according to nursing specifications. Dynamic time warping algorithm is a method for measuring the similarity between two time sequences, which calculates the difference between them by finding the best matching path between the two sequences. The window deviation is an index calculated based on this difference, which measures the degree of deviation of the nursing operation from the standard, and can provide a basis for subsequent abnormal intervention and nursing scheme adjustment.

[0132] Specifically, in the scheme execution monitoring module, the nursing operation is decomposed into a series of atomic events, which are arranged in the order of actual occurrence to form an atomic event sequence, while the standard sequence is predefined according to the nursing specification. The dynamic time warping algorithm calculates the minimum path difference between the two sequences by comparing them, and this difference reflects the degree of deviation between the actual operation and the standard operation. In the calculation process, the algorithm considers the characteristic values of each event, such as the time of the nursing operation, the type of operation, etc., and determines the path by comparing the differences between the actual execution characteristic values and the standard characteristic values. The calculation of the window deviation further considers the cumulative effect of such differences, and by setting a sensitivity adjustment coefficient, the weight of the differences is balanced, so as to obtain a quantitative index that can reflect the overall deviation degree of the nursing operation. The closer the value of this index to 1, the closer the nursing operation is to the standard operation; on the contrary, the smaller the value, the greater the deviation.

[0133] Preferably, when decomposing the nursing operation into an atomic event sequence, more event characteristics can be introduced, such as the duration of the operation, the performer of the operation, etc., to more comprehensively describe the process of the nursing operation. For the dynamic time warping algorithm, different distance measurement methods can be tried, such as Euclidean distance, Manhattan distance, etc., to calculate the difference between the characteristic values, in order to find the most suitable measurement method for the current data characteristics. When calculating the window deviation, the sensitivity adjustment coefficient can be adjusted according to the actual nursing scene and the tolerance degree of the deviation. For example, in a scene with high requirements for nursing quality, the sensitivity adjustment coefficient can be appropriately reduced, so that the window deviation can sensitively reflect even small deviations. In addition, the concept of time window can also be introduced to evaluate the nursing operation as a whole within a certain period of time, rather than only focusing on the deviation of a single event, so as to more accurately reflect the overall execution of the nursing operation.

[0134] In some embodiments, the nursing scheme reconstruction comprises:

[0135] setting the length of the sliding window and the time decay weight coefficient;

[0136] within the sliding window, accumulating the window deviation based on the time decay weight to obtain the cumulative deviation;

[0137] when the cumulative deviation and the current window deviation both exceed the preset threshold, triggering the dynamic decision module to reconstruct the nursing scheme.

[0138] Specifically, the step of triggering reconstruction by the abnormal intervention module comprises:

[0139] setting the length of the sliding window , the time decay constant is 1 hour;

[0140] The cumulative deviation within the calculation window is:

[0141] ,

[0142] wherein, represents an exponential decay weight, represents the deviation at the i-th time point, represents the cumulative deviation.

[0143] When the following conditions are met simultaneously, and the reconstruction of the scheme is triggered, wherein represents the square root of 6, and e represents the natural logarithm base;

[0144] It should be noted that the main function of the abnormal intervention module is to judge whether the nursing operation is abnormal by analyzing the execution deviation parameter, and to trigger the dynamic decision module to reconstruct the nursing scheme when necessary. This process involves setting the sliding window length and the time decay constant, calculating the cumulative deviation within the window, and judging whether the nursing scheme needs to be reconstructed according to the preset conditions. The sliding window length refers to the time range used to calculate the cumulative deviation, and the time decay constant is used to assign different weights to the deviations at different time points to reflect the decay characteristics of the deviation over time. When the cumulative deviation exceeds the preset threshold, it indicates that the nursing operation deviates greatly from the standard scheme, and the nursing scheme needs to be reconstructed to ensure the standardization and effectiveness of the nursing operation.

[0145] Specifically, the abnormal intervention module first sets the sliding window length to 6 time units (e.g. 6 hours) and the time decay constant to 24 hours. This means that when calculating the cumulative deviation, the deviations in the past 6 time units will be considered, and the influence of the deviation will gradually decrease over time. The calculation of the cumulative deviation is obtained by summing the deviations at each time point multiplied by the corresponding exponential decay weight. The formula for calculating the exponential decay weight is wherein represents the current time point, represents the past time point. When the cumulative deviation exceeds the preset threshold, for example, 0.7 times the square root of 6, and the average value of the deviation in the past 6 time units exceeds 0.9 times the decay coefficient, the reconstruction of the nursing scheme is triggered. These parameters can be adjusted according to the actual nursing scene and the tolerance of the deviation.

[0146] Preferably, when setting the sliding window length, it can be adjusted according to the complexity of the nursing operation and the requirement for real-time. For example, for nursing operations that require quick response, the sliding window length can be set shorter so that the abnormal situation can be detected faster. The time decay constant can also be adjusted according to the characteristics of the nursing operation, for example, for long-term nursing operations, the time decay constant can be appropriately increased to more smoothly reflect the changes in the deviation degree. When calculating the cumulative deviation degree, more statistical methods such as moving average or weighted moving average can be introduced to more accurately reflect the trend of the deviation degree. In addition, the threshold of the trigger condition can be dynamically adjusted according to the importance and risk level of the nursing operation, for example, in high-risk nursing operations, a lower threshold can be set to more sensitively detect abnormal situations.

[0147] In some embodiments, the multi-terminal collaborative module converts the nursing decision vector into a device instruction set, including the following steps:

[0148] Extracting an instruction feature vector from the nursing decision vector and obtaining an instruction delay time;

[0149] Calculating an urgency component based on the delay time, calculating an importance component based on the instruction feature vector and a preset importance coefficient, and weightedly summing to obtain an instruction priority;

[0150] If there are concurrent conflicting instructions, the instruction with shorter execution time or higher priority is selected according to the priority difference.

[0151] Specifically, the step of the multi-terminal collaborative module performing instruction arbitration includes:

[0152] Collecting an instruction feature vector and a delay time (unit: minutes), wherein, represents a 64-dimensional real number vector space;

[0153] Calculating an instruction priority:

[0154] ,

[0155] ,

[0156] ,

[0157] wherein, represents a learnable parameter vector, represents a preset importance coefficient, and softmax represents a normalized exponential function, represents an urgency component, represents an importance component, is the instruction priority;

[0158] Arbitrate conflicting instructions:

[0159] like , select execution time Instructions, where Indicates the estimated time of the i-th instruction; otherwise, the instruction with a larger P value is selected; is the priority of the ith instruction, is the priority of the jth instruction.

[0160] It should be noted that the core function of the multi-terminal collaboration module is to convert the nursing decision vector into a device instruction set, arbitrate and distribute the instructions of the terminal device, and at the same time feedback the device status to the dynamic decision module. This process involves collecting the instruction feature vector and delay time, calculating the instruction priority, and arbitrating conflicting instructions based on the priority. The instruction feature vector is a vector that describes the characteristics of the nursing decision vector and is used to determine the execution priority of the instruction. The delay time refers to the time interval between the generation of the instruction and the actual execution of the instruction by the device. The calculation of the instruction priority is based on the characteristics of the instruction and the preset importance coefficient, and is normalized by the softmax function to ensure the reasonable execution order of the instructions.

[0161] Specifically, the multi-terminal collaboration module first collects the instruction feature vector and delay time. The instruction feature vector is a 64-dimensional real number vector that contains the key feature information of the nursing decision vector. The delay time is measured in minutes and reflects the timeliness of instruction execution. When calculating the instruction priority, a basic priority is first calculated based on the delay time. The shorter the delay time, the higher the basic priority. Then, the instruction feature vector is weighted by a learnable parameter vector and a preset importance coefficient to obtain a comprehensive priority. The preset importance coefficient is a value between 0 and 1, which is used to adjust the weight of the instruction feature vector and delay time in the priority calculation. Finally, the softmax function is used to normalize the comprehensive priority of all instructions to obtain the final priority of each instruction. When arbitrating conflicting instructions, if the priority difference between the two instructions is less than 0.1, the instruction with the shorter execution time is selected; otherwise, the instruction with the higher priority is selected.

[0162] Preferably, when collecting the instruction feature vector, different feature extraction methods can be selected according to specific nursing scenarios and device types. For example, for complex nursing operations, a feature extraction network in deep learning can be introduced to more accurately extract instruction features. When calculating the instruction priority, different normalization methods such as max-min normalization or Z-score normalization can be tried to find the most suitable normalization method for the current data characteristics. For the preset importance coefficient, it can be dynamically adjusted according to the importance and urgency of the nursing operation. For example, in an emergency nursing scenario, the weight of the delay time can be appropriately increased to ensure that the instructions can be quickly executed. When arbitrating conflicting instructions, in addition to considering the priority and time consumption of the instructions, the current state and load of the device can also be introduced as a reference factor for arbitration to more reasonably allocate device resources and improve nursing efficiency.

[0163] In some embodiments, the step of encrypting the knowledge graph by the secure storage module comprises:

[0164] Step one: divide the graph into subgraphs , where represents the total number of fragments;

[0165] Step two: perform double encryption on each subgraph:

[0166] ,

[0167] where represents a 256-bit independent key, represents a bitwise XOR operation, represents the encrypted subgraph data, represents a symmetric encryption algorithm, represents a stream encryption algorithm.

[0168] Step three: generate an integrity verification tag:

[0169] ,

[0170] where represents data splicing operation, timestamp is the current Unix timestamp, represents the integrity verification tag, represents a 256-bit independent key, represents a message authentication code algorithm based on a hash function.

[0171] It should be noted that the core function of the secure storage module is to encrypt the feature topology graph and the decision knowledge graph, and generate an integrity verification tag to ensure the security and integrity of the data. This process involves splitting the graph into multiple subgraphs, double-encrypting each subgraph, and generating an integrity verification tag based on the encrypted data. The feature topology graph is a graph structure that describes the relationships between features, and the decision knowledge graph stores knowledge and rules related to nursing decisions. Double encryption is a method to enhance data security by using two independent keys to encrypt the data, ensuring that even if one key is cracked, the data remains secure. The integrity verification tag is used to detect whether the data has been tampered with during storage and transmission, ensuring that the stored data is complete and reliable.

[0172] Specifically, the secure storage module first splits the graph into 28 subgraphs, each containing a portion of the graph data. This splitting method can improve encryption efficiency and reduce the risk of a single subgraph being cracked. When double-encrypting each subgraph, two 256-bit independent keys are used, and different encryption algorithms are used to encrypt the subgraph. The first key uses the AES-256 algorithm for encryption, and the second key uses the SHA-256 algorithm to generate a hash value and then performs an XOR operation, thereby enhancing the complexity and security of the encryption. When generating the integrity verification tag, the encrypted subgraph data is concatenated with the current Unix timestamp, and then a hash value is generated through the SHA-256 algorithm. This hash value serves as the integrity verification tag for subsequent data integrity verification.

[0173] Preferably, when splitting the graph, the number of subgraphs can be dynamically adjusted according to the size and complexity of the graph. For example, for larger graphs, the number of subgraphs can be increased to improve encryption efficiency and security. When selecting encryption algorithms, more advanced encryption algorithms such as quantum encryption algorithms can be considered to address potential future security threats. For the generation of the integrity verification tag, more data elements such as device IDs or user IDs can be introduced to enhance the uniqueness and security of the tag. In addition, the encryption key and the generation algorithm of the integrity verification tag can be updated regularly to prevent security risks caused by long-term use of the same key or algorithm.

[0174] In some embodiments, the final nursing decision vector is generated by fusing multi-scale features through a gating mechanism, including:

[0175] Calculating the orthogonality loss between the multi-scale feature matrices, which is the square sum of the Frobenius norm of the transpose product of each scale feature matrix;

[0176] The total loss function for fusing multi-scale features through the gating mechanism includes the weighted sum of the cross-entropy loss and the orthogonality loss, where the weight of the orthogonality loss is 0.01.

[0177] Specifically, the gating fusion step applies an orthogonal constraint, which includes:

[0178] The orthogonal loss of the feature matrix is calculated as:

[0179] ,

[0180] where, represents the square of the matrix Frobenius norm, represents the transpose matrix of the 5x5x5 convolutional feature;

[0181] The total loss function is:

[0182] ,

[0183] where y represents the true nursing scheme label, represents the predicted result, and CrossEntropyLoss represents the cross-entropy loss function.

[0184] It should be noted that the gating fusion step applies an orthogonal constraint to optimize the performance of the deep spatio-temporal fusion network and ensure that the feature matrix maintains good structural characteristics during the fusion process. The orthogonal constraint is a mathematical constraint that requires the rows or columns of the feature matrix to be orthogonal to each other, i.e., their inner product is zero. This constraint can reduce the redundancy between features, enhance the independence and discrimination ability of features, and thus improve the generalization ability and stability of the model. The total loss function is a function used to measure the difference between the predicted result and the true label during the model training process. By adding the orthogonal constraint term to the loss function, the model parameters can be optimized while ensuring that the feature matrix satisfies the orthogonality requirement.

[0185] Specifically, the orthogonal constraint is calculated by multiplying the transpose of the feature matrix with itself to obtain a matrix, and then calculating the square of the Frobenius norm of this matrix. The Frobenius norm is a matrix norm that measures the square root of the sum of the squares of all elements in the matrix. In this system, the orthogonal losses of 5x5x5 convolutional features, 3x3x3 convolutional features, and 1x1x1 convolutional features are calculated respectively, and these losses are added to obtain the total orthogonal loss. The total loss function consists of two parts: the cross-entropy loss function and the orthogonal loss. The cross-entropy loss function is used to measure the difference between the predicted nursing decision vector and the true nursing scheme label, while the orthogonal loss is used to ensure the orthogonality of the feature matrix. In the total loss function, the weight of the cross-entropy loss function is 1, and the weight of the orthogonal loss is 0.01. This weight setting allows the model to focus on both prediction accuracy and feature matrix orthogonality during optimization.

[0186] Preferably, when calculating the orthogonal loss, the weight of the orthogonal loss can be dynamically adjusted according to the size of the feature matrix and the correlation between the features. For example, if the feature matrix is large and the correlation between the features is high, the weight of the orthogonal loss can be appropriately increased to more strictly impose the orthogonal constraint. In the total loss function, in addition to the cross-entropy loss function and the orthogonal loss, other types of loss functions such as regularization terms can also be introduced to further improve the generalization ability of the model. For example, an L2 regularization term can be added to penalize the weights of the model and prevent overfitting. In addition, different optimization algorithms such as the Adam optimization algorithm or the RMSprop optimization algorithm can also be tried to minimize the total loss function to improve the training efficiency and convergence speed of the model.

[0187] The above various embodiments of the present application have the following beneficial effects:

[0188] The nursing system obtains multi-source heterogeneous electronic medical record data from the hospital HIS system, wearable monitoring equipment, and nursing operation terminal through the heterogeneous data integration module, and outputs the data after cleaning and standardization, which can provide a unified format data basis for subsequent processing. The feature correlation module receives the data after cleaning and standardization to generate a spatio-temporal feature correlation matrix and store a feature topology graph, which can mine spatio-temporal correlation information in the data and provide a basis for precise nursing decision-making. The dynamic decision-making module processes the feature correlation matrix based on a deep spatio-temporal fusion network to generate a nursing decision vector, which can realize intelligent nursing decision-making. The scheme execution monitoring module collects nursing operation data streams in real time, calculates an execution deviation parameter, and feeds back, which can timely find deviations in nursing operations. The abnormal intervention module analyzes the execution deviation parameter to trigger the dynamic decision-making module to execute nursing scheme reconstruction, which can ensure dynamic adjustment and optimization of the nursing scheme. The multi-terminal collaborative module converts the nursing decision vector into a device instruction set and feeds back the device state, which can realize collaborative work of nursing equipment and improve nursing efficiency. The secure storage module encrypts and stores the feature topology graph and decision knowledge graph and provides model parameter version management, which can guarantee the security and reliability of the data.

[0189] In terms of data processing, the system uses a Markov chain model to estimate missing data and perform data normalization, which can improve the integrity and accuracy of the data. The feature correlation module can better capture the spatio-temporal features in the data by constructing a three-dimensional feature tensor and calculating a spatio-temporal attention weight matrix. The three-dimensional convolutional network performs dynamic kernel updating, which can enhance the model's ability to adapt to feature changes. The deep spatio-temporal fusion network processing procedure includes extracting physiological feature vectors and environmental parameter vectors, calculating cross-modal correlation weights, performing multi-scale convolution processing, and gating and fusing multi-scale features, which can achieve deep fusion and effective utilization of multi-modal data. The scheme execution monitoring module uses dynamic time warping and window deviation calculation methods to calculate the execution deviation, which can more accurately evaluate the execution of nursing operations. The abnormal intervention module calculates the cumulative deviation within the window by setting the sliding window length and time decay constant and triggers the scheme reconstruction, which can respond to abnormal situations in time and adjust the nursing scheme. The multi-terminal collaboration module considers the instruction priority and conflict resolution mechanism when arbitrating the execution instructions, which can ensure the reasonable execution of the instructions. The secure storage module uses double encryption and integrity verification tags to encrypt the knowledge graph, which can improve the security and integrity of the data. The gating and fusion step applies orthogonal constraints, which can optimize the performance of the model.

[0190] Further, the storage medium of the embodiments of the present application stores program instructions capable of implementing all the methods described above, wherein the program instructions can be stored in the storage medium in the form of a software product, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes, or a computer, a server, a mobile phone, a tablet, and other terminal devices.

[0191] The above description is only some of the preferred embodiments of the present application and an explanation of the technical principles applied. Those skilled in the art should understand that the scope of the application involved in the embodiments of the present application is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features can be replaced with technical features disclosed in the embodiments of the present application (but not limited to) having similar functions to form technical solutions.

Claims

1. An electronic medical record-based care system, characterized by, The method comprises the following steps: The heterogeneous data integration module is used for acquiring multi-source heterogeneous electronic medical record data from a hospital HIS system, a wearable monitoring device and a nursing operation terminal, and outputting the data to a feature correlation module after cleaning and standardizing the data; The feature correlation module is used for receiving the data after cleaning and standardizing and generating a space-time feature correlation matrix, and outputting the matrix to a dynamic decision module, and storing a feature topology graph to a secure storage module; The dynamic decision module is used for processing the space-time feature correlation matrix based on a deep space-time fusion network, generating a nursing decision vector, and distributing the vector to a scheme execution monitoring module and a multi-terminal collaborative module; The scheme execution monitoring module is used for collecting nursing operation data streams in real time, calculating an execution deviation parameter and feeding back to an abnormal intervention module and a dynamic decision module; The abnormal intervention module is used for analyzing the execution deviation parameter, and triggering the dynamic decision module to reconstruct a nursing scheme when the deviation exceeds a threshold value; The multi-terminal collaborative module is used for converting the nursing decision vector into a device instruction set, arbitrating conflicting instructions and then distributing to terminal devices, and feeding back device states to the dynamic decision module; The secure storage module is used for encrypting and storing the feature topology graph and the decision knowledge graph, and providing model parameter version management for the dynamic decision module; The feature correlation module performs the following steps: The data after cleaning and standardizing is constructed into a three-dimensional feature tensor according to time slices, ward areas and physiological index dimensions; A space-time attention weight matrix is calculated based on the three-dimensional feature tensor, as shown in the following formula; where Q i represents the query vector of the i-th time slice, K j represents the key vector of the j-th spatial location, W Q represents the query projection matrix, W K represents the key projection matrix, represents the scaling factor; the three-dimensional feature tensor T ∈ R 24×W×V where 24 represents the number of 24-hour time slices, W represents the total number of ward areas, and C represents the number of physiological indicator dimensions. The space-time attention weight matrix is spliced with the three-dimensional feature tensor, and then input into a three-dimensional convolution network for feature extraction; The deep space-time fusion network processing flow comprises: The physiological feature vector and the environmental parameter vector are extracted from the space-time feature correlation matrix respectively; The physiological feature vector and the environmental parameter vector are spliced, and a cross-modal correlation weight is calculated through a learnable parameter matrix; Based on the cross-modal correlation weight, the physiological feature vector is processed by multi-scale convolution, and features of different granularities are extracted by using large, medium and small convolution kernel sizes respectively; Multi-scale features are fused through a gating mechanism to generate a final nursing decision vector; When the deviation exceeds the threshold value, the dynamic decision module is triggered to reconstruct the nursing scheme, which comprises the following steps: A sliding window length and a time decay weight coefficient are set; In the sliding window, the window deviation is accumulated based on the time decay weight to obtain a cumulative deviation; When the cumulative deviation and the current window deviation both exceed a preset threshold value, the dynamic decision module is triggered to reconstruct the nursing scheme; The multi-terminal collaborative module converts the nursing decision vector into a device instruction set, which comprises the following steps: An instruction feature vector is extracted from the nursing decision vector, and an instruction delay time is obtained; An urgency component is calculated based on the delay time, an importance component is calculated based on the instruction feature vector and a preset importance coefficient, and a weighted sum is obtained to obtain an instruction priority; If there are concurrent conflicting instructions, the instruction with shorter execution time or higher priority is selected according to the priority difference; The multi-scale features are fused through a gating mechanism to generate a final nursing decision vector, which comprises the following steps: computing an orthogonality loss between the multi-scale feature matrices, the orthogonality loss being a Frobenius norm square sum of the transpose product of each scale feature matrix; the total loss function of the gated fusion includes a weighted sum of a cross-entropy loss and an orthogonality loss, wherein the weight of the orthogonality loss is 0.

01.

2. The system of claim 1, wherein, The data cleaning performed by the heterogeneous data integration module includes the following steps: Based on the Markov chain model, the missing data is estimated by the previous state transition probability, the historical mean and the Gaussian noise; Segmented nonlinear normalization is performed on the complete data, when the data value is greater than or equal to a preset reference threshold, it is mapped to the interval [0, 1]; when the data value is less than the reference threshold, it is mapped to the interval [0, 0.5].

3. The system of claim 1, wherein, The dynamic kernel update performed by the three-dimensional convolution network includes the following steps: Real-time monitoring of the feature change amount of the three-dimensional feature tensor; The convolution kernel weight is dynamically adjusted by the gated recurrent unit, wherein the current weight is determined by the one-dimensional vector of the previous weight and the feature change amount.

4. The system of claim 1, wherein, The execution deviation parameter calculated by the scheme execution monitoring module includes: The actual nursing operation is decomposed into an atomic event sequence, and is aligned with a preset standard event sequence; The minimum path difference between the actual sequence and the standard sequence is calculated by the dynamic time warping algorithm; Based on the minimum path difference and the sequence length, the window deviation is calculated by an exponential decay function, which is used to quantify the operation deviation.

Images