A smart bed management method and system

By performing characteristic analysis and disease prediction on the patient's brain wave, breathing and eye movement data, and reconstructing the bed allocation table, the problem that traditional bed management methods cannot respond to changes in the disease are solved, and efficient utilization of bed resources and patient care quality are achieved. promote.

CN118136213BActive Publication Date: 2024-11-26WUHAN YANHUANG INNOVATION TECH SERVICES CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410260355.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-07
Publication Date
2024-11-26
Estimated Expiration
2044-03-07

AI Technical Summary

Technical Problem

Traditional bed management methods are static and cannot respond to changes in patients' condition, resulting in severe patients failing to obtain necessary care in a timely manner or patients with mild patients occupying urgently needed bed resources.

Method used

By obtaining the patient's brain wave, breathing and eye movement data, performing characteristic analysis and condition prediction, and reconstructing the bed allocation table based on the condition priority to achieve dynamic bed management.

Benefits of technology

It ensures the efficient use of medical resources and the reasonable allocation of bed resources, can respond to changes in the disease and medical resources in real time, and improves the quality of patient care and the efficiency of bed resources utilization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118136213B_ABST
    Figure CN118136213B_ABST
Patent Text Reader

Abstract

The present invention relates to the medical field, and in particular to a smart bed management method and system. A smart bed management system includes: a vital sign monitoring data acquisition module, a feature analysis module, a condition prediction module, a condition priority analysis module, and a bed allocation table reconstruction module. The present invention collects the patient's biometric data, including brain wave, breathing and eye movement information, analyzes and predicts the patient's condition trend, and allocates beds based on the condition trend, ensuring that resources are first allocated to the patients who need them most, and can adapt to changes in the condition and changes in medical resource conditions in a timely manner.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the medical field, and in particular to a smart bed management method and system. Background Art

[0002] Traditional bed management methods are usually static, that is, once a patient is assigned to a bed, even if his or her condition changes, the bed allocation tends to remain unchanged until a new bed demand arises or the patient is discharged. This static bed management method ignores the real-time and dynamic nature of changes in the patient's condition and cannot ensure the optimal allocation of medical resources, which may result in critically ill patients not receiving the necessary care in a timely manner, or mildly ill patients occupying urgently needed bed resources. Summary of the invention

[0003] The present invention provides a bed management method that can flexibly respond to changes in the condition of the patient and changes in the status of medical resources.

[0004] A smart bed management method, comprising:

[0005] Step S1: obtaining the user's vital sign monitoring data within a preset period, the vital sign monitoring data including brain wave data, breathing data and eye movement data;

[0006] Step S2: performing feature analysis on the vital sign monitoring data to obtain vital sign feature data, wherein the vital sign feature data specifically includes brain wave feature data, breathing feature data and eye movement feature data corresponding to the vital sign monitoring data;

[0007] Step S3: Send the vital sign feature data to the disease prediction analysis model for processing, and output the disease prediction trend corresponding to the user;

[0008] Step S4: performing a disease priority analysis based on the disease prediction trend corresponding to the user;

[0009] Step S5: Reconstruct the bed allocation table according to the disease priority of all users to perform bed management.

[0010] The present invention can accurately collect the patient's vital signs information through advanced monitoring tools. Among these information, EEG data directly reflects the electrical activity of the brain, which is crucial for evaluating the progression of brain diseases. At the same time, respiratory data reveals the patient's physiological condition, especially for patients with damaged brain stem and respiratory control areas. This data provides an important reference. In addition, eye movement data, as a non-invasive indicator, can reflect the patient's cognitive state and the functional state of the nervous system. By efficiently processing and extracting features from these vital signs data, we not only simplify the data processing process, but also improve the accuracy and efficiency of subsequent analysis. This data dimensionality reduction processing plays a key role in processing a large amount of patient data and realizing fast dynamic bed allocation. By inputting these refined feature data into the disease prediction analysis model, medical staff can intuitively understand the patient's disease trend, thereby guiding the reasonable allocation of bed resources. Through a detailed analysis of disease priority, the hospital can ensure that resources are allocated to the patients in most need, while also achieving efficient use of medical resources and being able to respond to changes in disease and medical resources in real time.

[0011] As a preferred aspect, the specific content of step S2 is as follows:

[0012] Step S21: performing time domain analysis on the brain wave data to obtain brain wave time domain feature data;

[0013] Step S22: converting the brain wave data into the frequency domain through Fourier transform, and performing frequency domain analysis to obtain brain wave frequency domain feature data;

[0014] Step S23: performing wavelet transformation on the brain wave data, and then performing energy distribution analysis to output brain wave energy distribution data, and outputting brain wave wavelet feature data based on the brain wave energy distribution data;

[0015] Step S24: splicing the brain wave time domain feature data, the brain wave frequency domain feature data and the brain wave wavelet feature data to obtain brain wave feature data;

[0016] Step S25: extracting statistical features from the breathing data to obtain breathing feature data;

[0017] Step S26: extract statistical features from the eye movement data to obtain eye movement feature data.

[0018] The present invention extracts time domain features such as amplitude, variance, and peak value through time domain analysis of EEG data. These features can reveal the volatility and stability of EEG signals and provide basic information for understanding the user's EEG activity. Then, the EEG data is converted from the time domain to the frequency domain through Fourier transform, and features such as power spectrum density of different frequency bands (such as δ, θ, α, β, and γ waves) are extracted. The frequency domain features can help us understand the contribution and changes of different frequency components in EEG activity. Then, a more detailed time-frequency analysis of the EEG data is performed through wavelet transform to obtain information about EEG at different time and frequency scales, which can reveal instantaneous changes and local features in EEG activity. By calculating the energy distribution and wavelet entropy of wavelet coefficients, features describing the complexity and dynamic changes of EEG can be obtained. Statistical features such as respiratory rate, respiratory depth, and respiratory phase relationship are extracted to reflect the user's physiological state from another perspective. The extraction of eye movement statistical features can further reflect the situation of the user's brain. In summary, the present application can extract useful feature data from vital sign monitoring data through multi-dimensional feature analysis.

[0019] As a preferred aspect, the disease prediction analysis model in step S3 specifically includes a time series feature extraction layer, a tensor fusion layer, a feature fusion layer and a disease prediction trend output layer; wherein the time series feature extraction layer is established based on the Transformer model, and is used to perform time series analysis on the vital sign feature data to construct the vital sign time series features corresponding to the user, the vital sign time series features include brain wave time series feature data, breathing time series feature data and eye movement time series feature data, and the time series feature extraction layer includes a brain wave time series feature extraction unit, a breathing time series feature extraction unit and an eye movement time series feature extraction unit; the tensor fusion layer is used to perform tensor fusion on the vital sign time series features to construct a first vital sign fusion feature; the feature fusion layer is used to fuse the eye movement time series feature data with the first vital sign fusion feature based on the self-attention mechanism to construct a second vital sign fusion feature; the disease prediction trend output layer is established based on the BP neural network, and is used to output the disease prediction trend based on the second vital sign fusion feature.

[0020] As a preferred aspect, in step S3, the vital sign feature data is sent to the disease prediction analysis model for processing, and the disease prediction trend corresponding to the user is output, which specifically includes the following contents:

[0021] In the temporal feature extraction layer, the brain wave feature data within M preset cycles are spliced ​​from top to bottom in time sequence to construct a brain wave feature map, where M is the number of preset cycles; the brain wave feature map is sent to the brain wave temporal feature extraction unit for processing, and the brain wave temporal feature data is output; the breathing feature data within the M preset cycles are spliced ​​from top to bottom in time sequence to construct a breathing feature map, and then the breathing feature map is sent to the breathing temporal feature extraction unit for processing, and the breathing temporal feature data is output; the eye movement feature data within the M preset cycles are spliced ​​from top to bottom in time sequence to construct an eye movement feature map, and then the eye movement feature map is sent to the eye movement temporal feature extraction unit for processing, and the eye movement temporal feature data is output;

[0022] In the tensor fusion layer, the EEG timing feature data, the breathing timing feature data, and the eye movement timing feature data are spliced ​​according to the channels, and then a convolution operation is performed to obtain a vital sign feature map. The length, width, and height of the vital sign feature map are H, W, and C, respectively. Then, a global average pooling operation is performed on different dimensions of the vital sign feature map to obtain an H-dimensional feature map, a W-dimensional feature map, and a C-dimensional feature map, respectively. The size of the H-dimensional feature map is "1×W×C", the size of the W-dimensional feature map is "H×1×C", and the size of the C-dimensional feature map is "H×W×1". The H-dimensional feature map is processed by a softmax function to obtain an H-dimensional weight matrix, and the H-dimensional weight matrix is ​​multiplied by the data in the corresponding channel of the vital sign feature map. The W-dimensional feature map is processed by a softmax function to obtain a W-dimensional weight matrix, and the W-dimensional weight matrix is ​​multiplied by the data in the corresponding channel of the H-dimensional adjusted vital signs feature map, and then the W-dimensional weight matrix is ​​added to the H-dimensional adjusted vital signs feature map and then standardized to obtain the W-dimensional adjusted vital signs feature map; the C-dimensional feature map is processed by a softmax function to obtain a C-dimensional weight matrix, and the C-dimensional weight matrix is ​​multiplied by the data in the corresponding channel of the W-dimensional adjusted vital signs feature map, and then the W-dimensional adjusted vital signs feature map is added to the W-dimensional adjusted vital signs feature map and then standardized to obtain the first vital sign fusion feature;

[0023] In the feature fusion layer, an eye movement feature query matrix, an eye movement feature key matrix and an eye movement feature numerical matrix are constructed based on the eye movement temporal feature data, a vital sign query matrix, a vital sign key matrix and a vital sign numerical matrix are constructed based on the first vital sign fusion feature, and then a self-attention mechanism operation is performed based on the vital sign query matrix, the eye movement feature key matrix and the eye movement feature numerical matrix to obtain a vital sign-eye movement attention feature map; based on the eye movement feature query matrix, the vital sign key matrix and the vital sign numerical matrix, a self-attention mechanism operation is performed to obtain an eye movement-vital feature attention feature map; then an eye movement-vital feature attention query matrix is ​​constructed based on the eye movement-vital feature attention feature map, a vital sign-eye movement attention key matrix and a vital sign-eye movement attention numerical matrix are constructed based on the vital sign-eye movement attention feature map, and then a self-attention mechanism is performed based on the eye movement-vital feature attention query matrix, the vital sign-eye movement attention key matrix and the vital sign-eye movement attention numerical matrix to obtain a second vital sign fusion feature;

[0024] In the disease prediction trend output layer, the second vital sign fusion feature is sequentially passed through the input layer, hidden layer and output layer in the disease prediction trend output layer to obtain the disease prediction trend.

[0025] In the present invention, in the time series feature extraction layer, feature graphs are constructed by splicing feature data in different time periods, and the changes of vital signs in time series are learned; in the tensor fusion layer, global and local features are extracted through convolution operations and global average pooling; in the feature fusion layer, the correlation analysis between features is strengthened through the self-attention mechanism. Through these methods, the accuracy of user condition prediction is improved, providing an information basis for subsequent bed allocation.

[0026] As a preferred aspect, the training of the disease prediction analysis model specifically includes the following steps:

[0027] Acquire several vital sign feature data labeled by actual disease prediction trends, form a training set of all the vital sign feature data labeled by actual disease prediction trends, and send them to the disease prediction analysis model with parameter initialization for training, and calculate the loss value based on the actual disease prediction trend. When the loss value is within the preset range, output the trained disease prediction analysis model; if the loss value is not within the preset range, continue to train the disease prediction analysis model through the training set.

[0028] As a preferred aspect, step S4 includes the following specific contents:

[0029] Step S41: obtaining the current severity coefficient T and emergency coefficient F of the disease corresponding to the user;

[0030] Step S42: updating the severity coefficient T of the disease corresponding to the user by the following formula: T=T*θ, where θ is the predicted trend of the disease corresponding to the user;

[0031] Step S43: Calculate the user's disease priority δ by the following formula: δ=α 1 *T+α 2 *F, where α 1 and α 2 are the weight coefficients corresponding to the severity coefficient and the emergency coefficient, and satisfy α 1 +α 2 =1.

[0032] As a preferred aspect, step S5 specifically includes the following contents:

[0033] Step S51: Obtain user codes corresponding to all users and their corresponding disease priority levels δ, and arrange all user codes in descending order according to the corresponding disease priority levels δ to construct a table of user codes to be selected;

[0034] Step S52: Obtain bed codes of all beds and their corresponding bed evaluation coefficients A;

[0035] Step S53: Allocate the bed codes of all beds to the user codes. Under the constraints of the bed allocation coefficient Q and the bed change workload, the bed allocation coefficient Q is the product of the disease priority δ corresponding to the user code and the bed evaluation coefficient A corresponding to the bed code assigned to the user code. Through genetic algorithm optimization, the user codes and their corresponding bed codes under the optimal allocation method are combined into a bed allocation table to achieve reconstruction of the bed allocation table.

[0036] The present invention first sorts patients according to their condition priority to ensure that patients who need more urgent and intensive care can get beds first; secondly, the nursing effect of the bed is considered, based on the bed evaluation coefficient, to ensure that the patient can be allocated to the bed that best suits his condition; finally, the bed allocation is optimized through an algorithm to balance the condition priority and the bed nursing effect, while considering reducing the workload caused by bed changes, thereby improving the utilization efficiency of hospital bed resources and the quality of patient care.

[0037] A smart bed management system, comprising:

[0038] A vital sign monitoring data acquisition module is used to acquire vital sign monitoring data, which includes brain wave data, breathing data and eye movement data;

[0039] The feature analysis module is used to: perform feature analysis on the vital sign monitoring data to obtain vital sign feature data, wherein the vital sign feature data specifically includes brain wave feature data, breathing feature data and eye movement feature data corresponding to the vital sign monitoring data;

[0040] The disease prediction module is used to send the vital signs characteristic data into the disease prediction analysis model for processing and output the corresponding disease prediction trend of the user;

[0041] The disease priority analysis module is used to perform disease priority analysis based on the predicted disease trend of the user;

[0042] The bed allocation table reconstruction module is used to reconstruct the bed allocation table according to the disease condition priority of all users.

[0043] The present invention has the following advantages:

[0044] The present invention collects the patient's brain wave data, breathing data and eye movement data, and performs in-depth feature analysis on these vital sign monitoring data, including time domain analysis, frequency domain analysis and wavelet transform, so as to extract useful vital sign feature data; through these feature data, the system can build a disease prediction analysis model, which includes a time series feature extraction layer, a tensor fusion layer, a feature fusion layer and a disease prediction trend output layer. In the time series feature extraction layer, the system can capture the dynamic characteristics of vital signs changing over time; in the tensor fusion layer and the feature fusion layer, through advanced data processing technologies, such as the self-attention mechanism, the correlation analysis between different vital sign data is strengthened, and the sensitivity and accuracy of the model to the patient's disease change are improved; in addition, the system can also perform disease priority analysis based on the disease prediction trend, and then guide the reasonable allocation of bed resources. Through detailed priority analysis, the hospital can ensure that bed resources are allocated to the patients who need them most, while taking into account the nursing effect of the bed and the workload caused by the bed change, and realizing the efficient use of medical resources; in summary, the present invention provides a bed management method that can flexibly respond to changes in disease conditions and changes in medical resource conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a schematic diagram of the structure of the smart bed management system used in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0047] Example 1

[0048] A smart bed management method, comprising:

[0049] Step S1: within a preset period, which is manually set and can be 1 hour or half an hour, the user's vital sign monitoring data is obtained, and the vital sign monitoring data includes brain wave data, breathing data and eye movement data, wherein the brain wave data refers to the data obtained by monitoring the user through the data of the electroencephalogram, which can reflect the electrical activity of the brain, that is, reflect the changes in the user's brain disease, and is specifically stored in the form of voltage changes; the breathing data specifically refers to the data obtained by monitoring the user's breathing, for example, the chest and abdomen movement sensor is used to monitor the movement of the chest and abdomen to reflect the breathing depth and breathing frequency, and the nasal airflow sensor is used to monitor the changes in the nasal airflow, and the user analyzes the changes in breathing. These breathing data can reflect the changes in the user's brain disease to a certain extent, for example, brainstem damage or tumors may cause irregular breathing rhythm; the eye movement data refers to the data obtained by monitoring the user's eye movement, and it should be noted that since the user needs to be awake to obtain the eye movement data, the medical staff needs to obtain the eye movement data of the user in a timely manner at the preset time. The eye movement data can be obtained through an eye tracking device, including gaze point, eye movement trajectory and pupil changes;

[0050] Step S2: Perform feature analysis on the vital sign monitoring data to obtain vital sign feature data. The vital sign feature data specifically includes brain wave feature data, breathing feature data and eye movement feature data corresponding to the vital sign monitoring data. Since the vital sign monitoring data includes a large amount of redundant or invalid data, it will affect the subsequent brain disease condition analysis. Therefore, it is necessary to extract useful feature data from the vital sign monitoring data through feature analysis technology, so as to achieve data dimensionality reduction processing and speed up data processing efficiency.

[0051] Step S3: Send the vital sign feature data to the disease prediction analysis model for processing, and output the disease prediction trend corresponding to the user. The disease prediction trend can be roughly divided into aggravation, alleviation and balance. Specifically, the disease prediction trend is expressed as a number. When the disease prediction trend is greater than 1.2, it is considered to be aggravated. When the disease prediction trend is between 0.8 and 1.2, it is considered to be balanced. When the disease prediction trend is less than 0.8, it is considered to be alleviated. According to the disease prediction trend corresponding to the user, the priority of the user's disease can be adjusted, and then the bed allocation can be adjusted;

[0052] Step S4: Performing a disease priority analysis based on the predicted disease trend of the user, so as to adjust the disease priority of the user. In the actual hospital care process, due to limited medical resources, users with more serious diseases need to be allocated to better beds, such as advanced nursing beds;

[0053] Step S5: Reconstruct the bed allocation table according to the disease priority of all users to perform bed management.

[0054] The present application collects the vital signs data of patients through high-precision monitoring equipment, among which EEG data provides direct information on brain electrical activity and is an important indicator for evaluating changes in brain diseases. Respiration data reflects the patient's physiological state from another perspective, especially for patients with affected brainstem and respiratory control areas. Eye movement data provides non-invasive indicators of the patient's cognitive state and nervous system function. Then, by efficiently processing the large amount of collected vital signs data, the most valuable feature data is extracted, which not only reduces the complexity of data processing, but also improves the accuracy and efficiency of subsequent analysis. This dimensionality reduction processing of data plays a key role in processing large-scale patient data and realizing fast dynamic bed allocation. Then, by inputting the feature data into the disease prediction analysis model, medical staff can obtain an intuitive understanding of the patient's disease trend to guide the allocation of bed resources. Finally, by analyzing the disease priority, the hospital can ensure that resources are allocated to the patients who need them most, while also ensuring the efficient use of medical resources. Compared with the traditional static allocation method, this data-driven bed allocation method is more flexible and responsive, and can adapt to changes in disease conditions and changes in medical resource conditions in real time.

[0055] The specific content of step S2 is as follows:

[0056] Step S21: Perform time domain analysis on the EEG data to obtain EEG time domain feature data, including amplitude, variance, peak value, etc.;

[0057] Step S22: converting the brain wave data into the frequency domain through Fourier transform, and performing frequency domain analysis to obtain brain wave frequency domain characteristic data, including power spectrum density of different frequency bands, etc.;

[0058] Step S23: performing wavelet transform on the EEG data, the transform result includes the coefficients of the EEG data at different frequencies and positions, and then performing energy distribution analysis, the energy can be calculated by the square of the coefficient, outputting the EEG energy distribution data, and outputting the EEG wavelet feature data based on the EEG energy distribution data, the EEG wavelet feature data specifically includes wavelet entropy, etc.;

[0059] Step S24: splicing the brain wave time domain feature data, the brain wave frequency domain feature data and the brain wave wavelet feature data to obtain brain wave feature data;

[0060] Step S25: extracting statistical features from the respiratory data to obtain respiratory feature data, including respiratory rate, respiratory depth, and respiratory phase relationship, wherein the respiratory phase relationship is the ratio of the number of synchronous and asynchronous chest and abdominal breathing;

[0061] Step S26: extracting statistical features from the eye movement data to obtain eye movement feature data, including gaze point duration, eye movement track length, etc.;

[0062] This application extracts time domain features such as amplitude, variance, and peak value through time domain analysis of EEG data. These features can reveal the volatility and stability of EEG signals and provide basic information for understanding the user's EEG activity. Then, the EEG data is converted from the time domain to the frequency domain through Fourier transform, and features such as power spectrum density of different frequency bands (such as δ, θ, α, β, and γ waves) are extracted. Frequency domain features can help us understand the contribution and changes of different frequency components in EEG activity. Then, a more detailed time-frequency analysis of the EEG data is performed through wavelet transform to obtain information about EEG at different time and frequency scales, which can reveal instantaneous changes and local features in EEG activity. By calculating the energy distribution and wavelet entropy of wavelet coefficients, features describing the complexity and dynamic changes of EEG can be obtained. Extracting statistical features such as respiratory rate, respiratory depth, and respiratory phase relationship can reflect the user's physiological state from another perspective. Extracting statistical features of eye movement can further reflect the situation of the user's brain. In summary, this application can extract useful feature data from vital sign monitoring data through multi-dimensional feature analysis.

[0063] The disease prediction analysis model in step S3 specifically includes a time series feature extraction layer, a tensor fusion layer, a feature fusion layer and a disease prediction trend output layer; wherein the time series feature extraction layer is established based on the Transformer model, and is used to perform time series analysis on the vital sign feature data to construct the vital sign time series features corresponding to the user, the vital sign time series features include brain wave time series feature data, breathing time series feature data and eye movement time series feature data, and the time series feature extraction layer includes a brain wave time series feature extraction unit, a breathing time series feature extraction unit and an eye movement time series feature extraction unit; the tensor fusion layer is used to perform tensor fusion on the vital sign time series features to construct a first vital sign fusion feature, and tensor fusion can integrate feature data from different data sources, thereby improving the expressiveness of the data; the feature fusion layer is used to fuse the eye movement time series feature data with the first vital sign fusion feature based on the self-attention mechanism to construct a second vital sign fusion feature; the disease prediction trend output layer is established based on the BP neural network, and is used to output the disease prediction trend based on the second vital sign fusion feature.

[0064] In step S3, the vital sign feature data is sent to the disease prediction analysis model for processing, and the disease prediction trend corresponding to the user is output, which specifically includes the following contents:

[0065] In the temporal feature extraction layer, the brain wave feature data within M preset cycles are spliced ​​from top to bottom in chronological order to construct a brain wave feature map, and M is the number of preset cycles, which is determined by the size of the preset cycle, that is, calculated as one day. If the preset cycle is one hour, M is 24. Of course, it is not actually calculated as one day, but usually from 8 am to 8 pm; the brain wave feature map is sent to the brain wave temporal feature extraction unit for processing, and the brain wave temporal feature data is output; the breathing feature data within the M preset cycles are spliced ​​from top to bottom in chronological order to construct a breathing feature map, and then the breathing feature map is sent to the breathing temporal feature extraction unit for processing, and the breathing temporal feature data is output; the eye movement feature data within the M preset cycles are spliced ​​from top to bottom in chronological order to construct an eye movement feature map, and then the eye movement feature map is sent to the eye movement temporal feature extraction unit for processing, and the eye movement temporal feature data is output;

[0066] In the tensor fusion layer, the EEG temporal feature data, the respiration temporal feature data, and the eye movement temporal feature data are spliced ​​according to the channels, and then a convolution operation is performed to obtain a vital sign feature map. The length, width, and height of the vital sign feature map are H, W, and C, respectively, and the convolution kernel size can be "3×3". Then, a global average pooling operation is performed on different dimensions of the vital sign feature map, that is, a global average pooling operation is performed on the front, side, and top views of the vital sign feature map to obtain an H-dimensional feature map, a W-dimensional feature map, and a C-dimensional feature map, respectively. The size of the H-dimensional feature map is "1×W×C", the size of the W-dimensional feature map is "H×1×C", and the size of the C-dimensional feature map is "H×W×1". The H-dimensional feature map is processed by a softmax function to obtain an H-dimensional weight matrix, which is multiplied by the H-dimensional weight matrix and the data in the corresponding channel of the vital sign feature map, and then added to the vital sign feature map and standardized to obtain an H-dimensional adjusted vital sign feature. Figure; Process the W-dimensional feature map through the softmax function to obtain a W-dimensional weight matrix, multiply the W-dimensional weight matrix with the data in the corresponding channel of the H-dimensional adjusted vital signs feature map, and then add it to the H-dimensional adjusted vital signs feature map and standardize it to obtain the W-dimensional adjusted vital signs feature map; Process the C-dimensional feature map through the softmax function to obtain a C-dimensional weight matrix, multiply the C-dimensional weight matrix with the data in the corresponding channel of the W-dimensional adjusted vital signs feature map, and then add it to the W-dimensional adjusted vital signs feature map and standardize it to obtain the first vital sign fusion feature. The local features of the vital signs feature map can be obtained through the convolution operation, and the global average pooling operation of the three dimensions of the vital signs feature map can capture the global information of each dimension. The tensor fusion layer can further strengthen the learning of vital signs data and strengthen the subsequent prediction of brain diseases under the fusion of global features and local features.

[0067] In the feature fusion layer, an eye movement feature query matrix, an eye movement feature key matrix and an eye movement feature numerical matrix are constructed based on the eye movement temporal feature data. It should be noted that in the process of constructing the eye movement feature query matrix, the eye movement feature key matrix and the eye movement feature numerical matrix, the existing self-attention mechanism is adopted, that is, dot multiplication is performed with the corresponding query weight matrix, the key weight matrix and the numerical weight matrix, which will not be repeated here. The vital signs query matrix, the vital signs key matrix and the vital signs numerical matrix are constructed based on the first vital sign fusion feature, and then the self-attention mechanism operation is performed based on the vital signs query matrix, the eye movement feature key matrix and the eye movement feature numerical matrix. The self-attention mechanism operation also refers to the existing self-attention mechanism to obtain the vital signs-eye movement attention feature map; based on the eye movement feature query matrix, the vital signs key matrix and the vital signs numerical matrix, the self-attention mechanism operation is performed to obtain the eye movement-life feature attention map. force feature map; then, based on the eye movement-vital feature attention feature map, an eye movement-vital feature attention query matrix is ​​constructed; based on the vital sign-eye movement attention feature map, a vital sign-eye movement attention key matrix and a vital sign-eye movement attention numerical matrix are constructed; then, based on the eye movement-vital feature attention query matrix, the vital sign-eye movement attention key matrix and the vital sign-eye movement attention numerical matrix, a self-attention mechanism is executed to obtain a second vital sign fusion feature; since the features related to eye movement can better reflect the situation of brain disease, the first vital sign fusion feature is further strengthened by the related features of eye movement, which can strengthen the feature expression related to brain disease in the first vital sign fusion feature; and in the process of strengthening the first vital sign fusion feature by the related features of eye movement, cross self-attention strengthening is realized based on the self-attention mechanism, which can improve the correlation of different features in the first vital sign fusion feature;

[0068] In the disease prediction trend output layer, the second vital sign fusion feature is sequentially passed through the input layer, hidden layer and output layer in the disease prediction trend output layer to obtain the disease prediction trend.

[0069] In this application, in the time series feature extraction layer, feature data in different time periods are spliced ​​to construct feature maps, and the time series changes of vital signs are learned; in the tensor fusion layer, global and local features are extracted through convolution operations and global average pooling; in the feature fusion layer, the correlation analysis between features is strengthened through the self-attention mechanism. Through these methods, the accuracy of user condition prediction is improved, providing an information basis for subsequent bed allocation.

[0070] The training of the disease prediction analysis model specifically includes the following steps:

[0071] Obtain a number of vital sign feature data annotated by actual disease prediction trends. It should be noted that the actual disease prediction trends are annotated by professionals according to methods such as fuzzy comprehensive evaluation models, and can reflect the actual disease conditions of the corresponding users under different vital sign data. All vital sign feature data annotated by actual disease prediction trends are composed of a training set and sent to a disease prediction analysis model with parameter initialization for training. The loss value is calculated based on the actual disease prediction trend. When the loss value is within a preset range, the trained disease prediction analysis model is output, and the preset range is set by professionals. If the loss value is not within the preset range, the disease prediction analysis model continues to be trained through the training set.

[0072] Step S4 includes the following specific contents:

[0073] Step S41: Obtain the current severity coefficient T and emergency coefficient F of the user's corresponding disease. It should be noted that each brain disease has an initial severity coefficient and emergency coefficient, which are set in advance by professionals based on experience. The severity coefficient is used to characterize the impact of the corresponding brain disease on the user's life and health, and will change with the condition of the corresponding brain disease. The emergency coefficient is used to characterize the degree to which the corresponding brain disease needs to be treated immediately. Although some brain diseases are serious, they may not need to be treated immediately, and the emergency coefficient is a fixed parameter;

[0074] Step S42: updating the severity coefficient T of the disease corresponding to the user by the following formula: T=T*θ, where θ is the predicted trend of the disease corresponding to the user, and the updating process of the severity coefficient is the process in which the severity of the corresponding disease changes with the disease;

[0075] Step S43: Calculate the user's disease priority δ by the following formula: δ=α 1 *T+α 2 *F, where α 1 and α 2 are the weight coefficients corresponding to the severity coefficient and the emergency coefficient, and satisfy α 1 +α 2 =1, thereby realizing the disease priority analysis;

[0076] This application sets the severity coefficient and emergency coefficient according to the impact of each brain disease on life and health and the urgency of treatment, providing quantitative support for medical decision-making; and dynamically updates the severity coefficient according to changes in the condition, which can dynamically manage the patient's condition priority and achieve flexibility in bed allocation.

[0077] Step S5 specifically includes the following contents:

[0078] Step S51: Obtain user codes corresponding to all users and their corresponding disease priority levels δ, and arrange all user codes in descending order according to the corresponding disease priority levels δ to construct a table of user codes to be selected;

[0079] Step S52: Obtain bed codes of all beds and their corresponding bed evaluation coefficients A, where the bed evaluation coefficients are used to characterize the nursing effects of the corresponding beds, and the bed evaluation coefficients of beds of the same grade are consistent;

[0080] Step S53: Allocate the bed codes of all beds to the user codes, and then under the constraints of the bed allocation coefficient Q and the bed change workload, the bed allocation coefficient Q is the product of the disease priority δ corresponding to the user code and the bed evaluation coefficient A corresponding to the bed code assigned to the user code. The bed change workload can be measured by the distance of bed changes. Through genetic algorithm optimization, the user codes and their corresponding bed codes under the optimal allocation method are combined into a bed allocation table to achieve reconstruction of the bed allocation table.

[0081] In the present application, the bed evaluation coefficient corresponding to the bed code is allocated to the user code corresponding to the disease priority which is also arranged from large to small in order. However, since there are beds of the same level, there are multiple allocation methods. Therefore, the user codes under various allocation methods and their corresponding bed codes are used as the initial population, which are selected, reorganized and mutated through a genetic algorithm. During this period, the bed allocation coefficient and the bed replacement workload are used as fitness. Finally, the user codes under the optimal allocation method and their corresponding bed codes are combined into a bed allocation table, which can realize dynamic bed management according to the user's disease condition and ensure that the workload of replacement is minimized when the bed is adjusted.

[0082] Example 2

[0083] A smart bed management system, see Figure 1 ,include:

[0084] A vital sign monitoring data acquisition module is used to acquire vital sign monitoring data, which includes brain wave data, breathing data and eye movement data;

[0085] The feature analysis module is used to: perform feature analysis on the vital sign monitoring data to obtain vital sign feature data, wherein the vital sign feature data specifically includes brain wave feature data, breathing feature data and eye movement feature data corresponding to the vital sign monitoring data;

[0086] The disease prediction module is used to send the vital signs characteristic data into the disease prediction analysis model for processing and output the corresponding disease prediction trend of the user;

[0087] The disease priority analysis module is used to perform disease priority analysis based on the predicted disease trend of the user;

[0088] The bed allocation table reconstruction module is used to reconstruct the bed allocation table according to the disease condition priority of all users.

[0089] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A smart bed management method, characterized in that: include: Step S1: obtaining the user's vital sign monitoring data within a preset period, the vital sign monitoring data including brain wave data, breathing data and eye movement data; Step S2: performing feature analysis on the vital sign monitoring data to obtain vital sign feature data, wherein the vital sign feature data specifically includes brain wave feature data, breathing feature data and eye movement feature data corresponding to the vital sign monitoring data; Step S3: Send the vital sign feature data to the disease prediction analysis model for processing, and output the disease prediction trend corresponding to the user; Step S4: performing a disease priority analysis based on the disease prediction trend corresponding to the user; Step S5: Reconstruct the bed allocation table according to the disease priority of all users to perform bed management; The specific content of step S2 is as follows: Step S21: performing time domain analysis on the brain wave data to obtain brain wave time domain feature data; Step S22: converting the brain wave data into the frequency domain through Fourier transform, and performing frequency domain analysis to obtain brain wave frequency domain feature data; Step S23: performing wavelet transformation on the brain wave data, and then performing energy distribution analysis to output brain wave energy distribution data, and outputting brain wave wavelet feature data based on the brain wave energy distribution data; Step S24: splicing the brain wave time domain feature data, the brain wave frequency domain feature data and the brain wave wavelet feature data to obtain brain wave feature data; Step S25: extracting statistical features from the breathing data to obtain breathing feature data; Step S26: extracting statistical features from the eye movement data to obtain eye movement feature data; The disease prediction analysis model in step S3 specifically includes a time series feature extraction layer, a tensor fusion layer, a feature fusion layer and a disease prediction trend output layer; wherein the time series feature extraction layer is established based on the Transformer model, and is used to perform time series analysis on the vital sign feature data to construct the vital sign time series features corresponding to the user, the vital sign time series features include brain wave time series feature data, breathing time series feature data and eye movement time series feature data, and the time series feature extraction layer includes a brain wave time series feature extraction unit, a breathing time series feature extraction unit and an eye movement time series feature extraction unit; the tensor fusion layer is used to perform tensor fusion on the vital sign time series features to construct the first vital sign fusion feature; feature The fusion layer is used to fuse the eye movement temporal feature data with the first vital sign fusion feature based on the self-attention mechanism to construct the second vital sign fusion feature; The disease prediction trend output layer is established based on the BP neural network, which is used to output the disease prediction trend based on the second vital sign fusion feature; In step S3, the vital sign feature data is sent to the disease prediction analysis model for processing, and the disease prediction trend corresponding to the user is output, which specifically includes the following contents: In the time series feature extraction layer, the brain wave feature data within M preset cycles are spliced ​​from top to bottom in chronological order to construct a brain wave feature map, where M is the number of preset cycles; The brain wave characteristic graph is sent to the brain wave time series characteristic extraction unit for processing, and the brain wave time series characteristic data is output; The respiratory characteristic data within the M preset periods are spliced ​​from top to bottom in chronological order to construct a respiratory characteristic graph, and then the respiratory characteristic graph is sent to the respiratory timing characteristic extraction unit for processing to output the respiratory timing characteristic data; The eye movement feature data in the M preset periods are spliced ​​from top to bottom in chronological order to construct an eye movement feature map, and then the eye movement feature map is sent to the eye movement temporal feature extraction unit for processing to output the eye movement temporal feature data; In the tensor fusion layer, the EEG timing feature data, the breathing timing feature data, and the eye movement timing feature data are spliced ​​according to the channels, and then a convolution operation is performed to obtain a vital sign feature map. The length, width, and height of the vital sign feature map are H, W, and C, respectively. Then, a global average pooling operation is performed on different dimensions of the vital sign feature map to obtain an H-dimensional feature map, a W-dimensional feature map, and a C-dimensional feature map, respectively. The size of the H-dimensional feature map is "1×W×C", the size of the W-dimensional feature map is "H×1×C", and the size of the C-dimensional feature map is "H×W×1". The H-dimensional feature map is processed by a softmax function to obtain an H-dimensional weight matrix, and the H-dimensional weight matrix is ​​multiplied by the data in the corresponding channel of the vital sign feature map. The W-dimensional feature map is processed by a softmax function to obtain a W-dimensional weight matrix, and the W-dimensional weight matrix is ​​multiplied by the data in the corresponding channel of the H-dimensional adjusted vital signs feature map, and then the W-dimensional weight matrix is ​​added to the H-dimensional adjusted vital signs feature map and then standardized to obtain the W-dimensional adjusted vital signs feature map; the C-dimensional feature map is processed by a softmax function to obtain a C-dimensional weight matrix, and the C-dimensional weight matrix is ​​multiplied by the data in the corresponding channel of the W-dimensional adjusted vital signs feature map, and then the W-dimensional adjusted vital signs feature map is added to the W-dimensional adjusted vital signs feature map and then standardized to obtain the first vital sign fusion feature; In the feature fusion layer, an eye movement feature query matrix, an eye movement feature key matrix and an eye movement feature numerical matrix are constructed based on the eye movement temporal feature data, a vital sign query matrix, a vital sign key matrix and a vital sign numerical matrix are constructed based on the first vital sign fusion feature, and then a self-attention mechanism operation is performed based on the vital sign query matrix, the eye movement feature key matrix and the eye movement feature numerical matrix to obtain a vital sign-eye movement attention feature map; based on the eye movement feature query matrix, the vital sign key matrix and the vital sign numerical matrix, a self-attention mechanism operation is performed to obtain an eye movement-vital feature attention feature map; then an eye movement-vital feature attention query matrix is ​​constructed based on the eye movement-vital feature attention feature map, a vital sign-eye movement attention key matrix and a vital sign-eye movement attention numerical matrix are constructed based on the vital sign-eye movement attention feature map, and then a self-attention mechanism is performed based on the eye movement-vital feature attention query matrix, the vital sign-eye movement attention key matrix and the vital sign-eye movement attention numerical matrix to obtain a second vital sign fusion feature; In the disease prediction trend output layer, the second vital sign fusion feature is sequentially passed through the input layer, hidden layer and output layer in the disease prediction trend output layer to obtain the disease prediction trend.

2. A smart bed management method according to claim 1, characterized in that: The training of the disease prediction analysis model specifically includes the following steps: Acquire several vital sign feature data labeled by actual disease prediction trends, form a training set of all the vital sign feature data labeled by actual disease prediction trends, and send them to the disease prediction analysis model with parameter initialization for training, and calculate the loss value based on the actual disease prediction trend. When the loss value is within the preset range, output the trained disease prediction analysis model; if the loss value is not within the preset range, continue to train the disease prediction analysis model through the training set.

3. A smart bed management method according to claim 2, characterized in that: Step S4 includes the following specific contents: Step S41: obtaining the current severity coefficient T and emergency coefficient F of the disease corresponding to the user; Step S42: updating the severity coefficient T of the disease corresponding to the user by the following formula: T=T*θ, where θ is the predicted trend of the disease corresponding to the user; Step S43: Calculate the user's corresponding disease priority δ by the following formula: δ=α1*T+α2*F, where α1 and α2 are weight coefficients corresponding to the severity coefficient and the urgency coefficient, respectively, and α1+α2=1.

4. A smart bed management method according to claim 3, characterized in that: Step S5 specifically includes the following contents: Step S51: Obtain user codes corresponding to all users and their corresponding disease priority levels δ, and arrange all user codes in descending order according to the corresponding disease priority levels δ to construct a table of user codes to be selected; Step S52: Obtain bed codes of all beds and their corresponding bed evaluation coefficients A; Step S53: Allocate the bed codes of all beds to the user codes. Under the constraints of the bed allocation coefficient Q and the bed change workload, the bed allocation coefficient Q is the product of the disease priority δ corresponding to the user code and the bed evaluation coefficient A corresponding to the bed code assigned to the user code. Through genetic algorithm optimization, the user codes and their corresponding bed codes under the optimal allocation method are combined into a bed allocation table to achieve reconstruction of the bed allocation table.

5. A smart bed management system, characterized in that: The system applies a smart bed management method according to any one of claims 1 to 4, including: A vital sign monitoring data acquisition module is used to acquire vital sign monitoring data, which includes brain wave data, breathing data and eye movement data; The feature analysis module is used to: perform feature analysis on the vital sign monitoring data to obtain vital sign feature data, wherein the vital sign feature data specifically includes brain wave feature data, breathing feature data and eye movement feature data corresponding to the vital sign monitoring data; The disease prediction module is used to send the vital signs characteristic data into the disease prediction analysis model for processing and output the corresponding disease prediction trend of the user; The disease priority analysis module is used to perform disease priority analysis based on the predicted disease trend of the user; The bed allocation table reconstruction module is used to reconstruct the bed allocation table according to the disease condition priority of all users.