Anesthesia recovery period patient grading intelligent evaluation system and method

Abstract

The invention discloses a grading intelligent evaluation system and method for patients in an anesthesia recovery period, belongs to the technical field of anesthesia, and solves the problems that subtle changes of numerous physiological parameters of the patients are difficult to comprehensively and continuously monitor by adopting a traditional evaluation mode and depending on self-evaluation of indexes of a monitor and the like at present; the method comprises the following steps: collecting physical sign monitoring data of a patient in real time based on a sensor and an Internet of Things technology, carrying out multi-dimensional parallel evaluation analysis on the physical sign monitoring data of the patient by an anesthesia recovery period patient evaluation model, and outputting an evaluation analysis result of the patient. Judging whether the real-time state evaluation level exceeds a preset patient risk threshold value or not; according to the method, the anesthesia recovery period patient evaluation model is used for carrying out multi-dimensional parallel evaluation analysis on the physical sign monitoring data of the patient, the nursing grading requirement of the patient is judged in an intelligent mode, objectivity and scientificity of nursing grading decision making are ensured, nursing quality is improved, complication risks are reduced, and meanwhile the comfort degree and satisfaction degree of the patient are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of anesthesia, and particularly relates to a patient grading intelligent evaluation system and method in the anesthesia recovery period. BACKGROUND

[0002] With the continuous deepening and expansion of the connotation of anesthesia medical services, the range of patients admitted to the anesthesia recovery period is increasingly extensive. The complexity of patient surgical and anesthesia plans has significantly increased, and the age span of patients has also increased, with an increasing number of combined diseases. These factors together result in greater differentiation in postoperative anesthesia status, disease severity, and complication occurrence among patients in the anesthesia recovery period.

[0003] The service range of the post anesthesia care unit (PACU) has gradually expanded from the original tertiary hospitals to secondary hospitals. Patients admitted to different level hospitals show great specificity in terms of disease types. In addition, the service content of the post anesthesia care unit has gradually expanded from traditional indoor anesthesia to cover the operating room, post anesthesia care unit, and post-anesthesia intensive care monitoring ward, as well as outdoor anesthesia (such as painless gastrointestinal endoscopy, bronchoscopy room, etc.). Therefore, there is significant heterogeneity among patients admitted to different post anesthesia care units. More complicatedly, all patients immediately transferred to the post anesthesia care unit have different immediate conditions, including whether they have a tracheal tube, an arterial catheter, a deep vein catheter, and whether they are conscious or not. These factors all have different effects on the treatment and intervention needs of the anesthesia recovery period.

[0004] However, the traditional anesthesia recovery period patient condition evaluation system largely relies on the subjective experience and limited bedside observation of medical personnel. This mode has many limitations:

[0005] Firstly, the manual evaluation method is difficult to comprehensively and continuously monitor the subtle changes in a patient's numerous physiological parameters. Due to the limitations of human energy and attention, it is easy to miss some key information during long-term monitoring, which may affect the accuracy of patient classification and have an adverse impact on the patient's treatment plan and recovery process.

[0006] Secondly, the differences in professional level and experience of different medical personnel also make the evaluation results highly subjective. Even when faced with the same patient and condition, different medical personnel may give different evaluation results based on their own experience and judgment, which not only lacks a unified standard but also greatly reduces the credibility of the evaluation results.

[0007] In addition, the conventional method also lacks the deep mining and analysis ability of a large amount of historical data and multi-source data, and cannot fully utilize the existing data resources. In view of the above problems, the present application provides a patient grading intelligent evaluation system and method in the anesthesia recovery period. SUMMARY

[0008] The present application aims at the deficiencies of the prior art, and provides a patient grading intelligent evaluation system and method in the anesthesia recovery period, which solves the problem that the conventional evaluation method is difficult to comprehensively and continuously monitor the subtle changes of a large number of physiological parameters of the patient, and the evaluation result has a large subjective nature.

[0009] The present application is implemented as follows: a patient grading intelligent evaluation method in the anesthesia recovery period, the method comprising:

[0010] Real-time acquisition of patient physical sign monitoring data based on sensors and Internet of Things technology, pre-processing of the patient physical sign monitoring data, and uploading of the pre-processed patient physical sign monitoring data to a historical database;

[0011] Pre-constructing an anesthesia recovery period patient evaluation model based on a convolutional neural network, traversing the historical database, grabbing a modeling sample set from the historical database, dividing the modeling sample set into an identifiable data format, and training the anesthesia recovery period patient evaluation model using the modeling sample set;

[0012] Obtaining the pre-processed patient physical sign monitoring data, multi-dimensional parallel evaluation and analysis of the patient physical sign monitoring data by the anesthesia recovery period patient evaluation model, outputting the evaluation and analysis result of the patient, and the evaluation and analysis result including the real-time state evaluation grade, the nursing demand grade, and the anesthesia recovery period patient grading;

[0013] Loading the evaluation and analysis result, judging whether the real-time state evaluation grade exceeds the preset patient risk threshold, triggering an early warning signal if the preset patient risk threshold is exceeded, and triggering a continuous monitoring signal if the preset patient risk threshold is not exceeded;

[0014] In response to the real-time state evaluation grade judgment result, the early warning signal / continuous monitoring signal, the nursing demand grade, and the anesthesia recovery period patient grading are visualized and presented, and the nursing demand grade and the anesthesia recovery period patient grading are pushed.

[0015] Preferably, the pre-processing method of the patient physical sign monitoring data comprises:

[0016] The physiological parameters of the patient's signs are acquired by a sensor distribution, the sensor including a heart rate sensor, a blood pressure sensor, a blood oxygen saturation sensor, a motion sensor, a monitoring camera, and a link is established with a hospital information system, the preoperative monitoring information, intraoperative monitoring information and postoperative monitoring information are extracted from the hospital information system, and the preoperative monitoring information, intraoperative monitoring information and postoperative monitoring information are combined with the physiological parameters of the patient's signs to obtain the combined physiological parameters of the patient's signs;

[0017] The patient's basic information is acquired based on the interaction between the Internet of Things technology and the hospital information system containing the patient's basic information;

[0018] The combined physiological parameters of the patient's signs are loaded, the physiological parameters of the patient's signs are filtered and denoised, and the physiological parameters of the patient's signs are cut based on the number of peaks of the physiological parameters of the patient's signs to cut at least one parameter time window;

[0019] The parameter time window containing the incidental event is extracted combined with the probability driving mechanism, and the parameters in the parameter time window containing the incidental event are used as the incidental physiological parameters, and the mean value of the parameters in the parameter time window not containing the incidental event is used as the virtual physiological parameters of the patient's signs, and the virtual physiological parameters of the patient's signs and the incidental physiological parameters are integrated as the denoised physiological parameters of the patient's signs;

[0020] The denoised physiological parameters of the patient's signs and the patient's basic information are acquired, the physiological parameters of the patient's signs and the patient's basic information are aligned based on the timestamp alignment algorithm, the physiological parameters of the patient's signs and the patient's basic information are fused by the federated feature fusion method, and the preprocessed patient's sign monitoring data is output.

[0021] Preferably, the method of filtering and denoising the physiological parameters of the patient's signs comprises:

[0022] The similarity of the physiological parameters of the patient's signs is detected based on the template matching technology, the physiological parameters of the patient's signs with a similarity coefficient exceeding a similarity threshold are screened through the preset similarity threshold, and the physiological parameters of the patient's signs exceeding the similarity threshold are regarded as effective physiological parameters of the patient's signs;

[0023] The effective physiological parameters of the patient's signs are loaded, the physiological parameters of the patient's signs are decomposed and reconstructed based on the CEEMD method, the physiological parameters of the patient's signs are divided into at least one IMF component, and the permutation entropy of the physiological parameters of the patient's signs is calculated;

[0024] The IMF components of the physiological parameters of the patient's signs are divided into low-frequency signals, trend signals and high-frequency signals based on a preset entropy value interval;

[0025] The divided low-frequency signal, trend signal and high-frequency signal are loaded, the low-frequency signal is filtered by a Butterworth low-pass filter, the trend signal is filtered by a Savitzky-Golay filter, and the high-frequency signal is filtered by a notch filter.

[0026] Preferably, the parameter time window method containing the contingency event extracted by the combination probability driving mechanism comprises:

[0027] The subjective weights of different types of parameters are determined based on an expert consultation method, the objective weights of different types of parameters are determined by an entropy weight method, a weight decision matrix containing the subjective weights and the objective weights is constructed by combining a TOPSIS model, and the parameter combination weights are obtained based on the weight decision matrix;

[0028] Based on the parameter combination weights, the parameter variation threshold and the probability integral threshold of the preset probability driving mechanism are determined, at least one set of parameter time windows is traversed, the parameter variation coefficient in the parameter time window is calculated based on the probability driving mechanism;

[0029] It is judged whether the parameter variation coefficient in the parameter time window exceeds the preset parameter variation threshold, if the parameter variation threshold exceeds the preset parameter variation threshold, the corresponding parameter variation coefficient is placed in the probability integral space by the probability driving mechanism, and if the parameter variation threshold exceeds the preset parameter variation threshold, the parameter variation coefficient is set to 0;

[0030] When the parameter variation coefficient in the probability integral space exceeds the preset probability integral threshold, the probability driving mechanism triggers the extraction mechanism to extract the parameter time window containing the contingency event;

[0031] The parameter mean value in the parameter time window is used as the virtual sign physiological parameter for the parameter time window not containing the contingency event.

[0032] Preferably, the method for fusing the patient sign physiological parameters and the patient basic information by the federated feature fusion method comprises:

[0033] The patient sign physiological parameters and the patient basic information are obtained, and the parameter types corresponding to the patient ID are aligned by using encrypted hash matching;

[0034] The parameter types are screened based on the parameter combination weights, and the screened patient sign physiological parameters and patient basic information are processed by federated feature alignment;

[0035] The patient sign physiological parameters and patient basic information processed by the federated feature alignment are processed by federated transfer learning by using a FedBN adjustment model, an encrypted exchange attention matrix is constructed, the patient sign physiological parameters and patient basic information are fused by using the federated feature fusion method combined with the encrypted exchange attention matrix, and the preprocessed patient sign monitoring data is output.

[0036] Preferably, the method for training the anesthesia recovery period patient evaluation model by using the modeling sample set comprises the following steps:

[0037] Load the modeling sample set, preprocess the modeling sample set by using the mixed enhancement method, and divide the modeling sample set into a training set and a test set;

[0038] Load the pre-constructed anesthesia recovery period patient evaluation model, and preset the parameter optimizer, training round, data batch, hyperparameter, and early stopping condition of the anesthesia recovery period patient evaluation model;

[0039] Obtain the training set, iteratively train the pre-constructed anesthesia recovery period patient evaluation model by using the multi-stage transfer learning strategy, perform Fourier transform on the training samples in the training set, perform Fourier frequency analysis on the sample features after the Fourier transform, filter the sample features higher than the frequency threshold based on the preset frequency threshold, perform pruning processing on the sample features lower than the frequency threshold, and output the lightweight converged anesthesia recovery period patient evaluation model;

[0040] Obtain the test set, execute the anesthesia recovery period patient evaluation model by using the test set as the input, output the test result, and determine whether the error item of the test result and the true result meets the error threshold;

[0041] If the error item of the test result and the true result meets the error threshold, output the converged anesthesia recovery period patient evaluation model;

[0042] If the error item of the test result and the true result does not meet the error threshold, adjust the model hyperparameter by using the RAdam optimizer, and continue to iteratively train the pre-constructed anesthesia recovery period patient evaluation model by using the training set and the multi-stage transfer learning strategy.

[0043] Preferably, the anesthesia recovery patient evaluation model takes a convolutional neural network as an initial model, and introduces an input layer and an output layer in the initial model, the input layer is connected with the convolutional neural network, the convolutional neural network is connected with the output layer, the convolutional neural network includes three convolutional blocks, a max pooling layer, and a global pooling layer, the global pooling layer of the convolutional neural network is frozen, a residual connection layer is used to replace the global pooling layer, the residual connection layer is embedded in a residual connection network ResNet, an Embedding processing layer and a 3D Conv layer are introduced between the three convolutional blocks and the max pooling layer, the loss function formula of the improved convolutional neural network is a cross-entropy loss function, a multi-task output layer is embedded between the convolutional neural network and the output layer, the multi-task output layer is used for evaluating and outputting the real-time state evaluation level of the patient, the nursing demand level, and the anesthesia recovery patient classification, the multi-task output layer includes a state evaluation block, a nursing evaluation block, and a PPL prediction block (PACU Patient Level, PPL), the activation function of the state evaluation block is a Softmax function, the activation function of the nursing evaluation block is a Sigmoid function, and the activation function of the PPL prediction block is a Linear function, the state evaluation block is a spiking neural network, and the spiking neural network introduces a decision-level fusion algorithm, the nursing evaluation block includes three CFCs layers and a fully connected layer, and the PPL prediction block is an LSTM model, and the LSTM model introduces an L-M algorithm.

[0044] Preferably, the anesthesia recovery patient evaluation model uses a multi-dimensional parallel evaluation and analysis method for patient sign monitoring data, which includes:

[0045] The preprocessed patient sign monitoring data is obtained, the input layer of the anesthesia recovery patient evaluation model performs step-up encoding processing on the patient sign monitoring data to obtain a step-up set of encoded data;

[0046] The step-up set of data is loaded and input into the convolutional neural network, the convolutional neural network identifies the parameter type, and respectively performs convolution fusion processing on the step-up set of data based on the parameter type, and outputs a feature extraction set;

[0047] The pulse timestamp, pulse sequence entropy, and feature self-correlation in the feature extraction set are counted based on the spiking neural network, the feature fusion vector of the feature extraction set is calculated based on the pulse timestamp, pulse sequence entropy, and feature self-correlation, the feature fusion vector is weighted and fused based on the decision-level fusion algorithm, and the real-time state evaluation level of the patient is output;

[0048] The real-time state evaluation level of the patient is obtained, the nursing evaluation block approximates the analytical solution of the patient's nursing demand based on the CFCs algorithm combined with the real-time state evaluation level, and takes the analytical solution of the patient's nursing demand as the nursing demand level of the patient;

[0049] The real-time state evaluation level of the patient and the nursing demand level of the patient are taken as inputs of a PPL prediction block, a patient classification target function in the anesthesia recovery period is constructed, the PPL prediction block calculates the patient classification target function in the anesthesia recovery period based on an L-M algorithm, and outputs the patient classification in the anesthesia recovery period.

[0050] The real-time state evaluation level of the patient, the nursing demand level of the patient and the patient classification in the anesthesia recovery period are obtained, and the real-time state evaluation level of the patient, the nursing demand level of the patient and the patient classification in the anesthesia recovery period are integrated as the evaluation analysis result of the patient.

[0051] In another aspect, the present application also provides an intelligent evaluation system for patient classification in the anesthesia recovery period, which comprises:

[0052] A data acquisition module acquires patient sign monitoring data in real time based on sensors and Internet of Things technology, pre-processes the patient sign monitoring data, and uploads the pre-processed patient sign monitoring data to a historical database;

[0053] An intelligent evaluation module is configured to construct a patient evaluation model in the anesthesia recovery period based on a convolutional neural network, and to acquire the pre-processed patient sign monitoring data, wherein the patient evaluation model in the anesthesia recovery period performs multi-dimensional parallel evaluation analysis on the patient sign monitoring data, and outputs the evaluation analysis result of the patient, which includes the real-time state evaluation level of the patient, the nursing demand level of the patient and the patient classification in the anesthesia recovery period;

[0054] An early warning and alarm module is configured to load the evaluation analysis result, to determine whether the real-time state evaluation level exceeds a preset patient risk threshold, to trigger an early warning signal if the preset patient risk threshold is exceeded, and to trigger a continuous monitoring signal if the preset patient risk threshold is not exceeded;

[0055] A data visualization module is configured to respond to the real-time state evaluation level determination result, to visually present the early warning signal / continuous monitoring signal, the nursing demand level and the patient classification in the anesthesia recovery period, and to push the nursing demand level and the patient classification in the anesthesia recovery period.

[0056] Preferably, the data acquisition module comprises:

[0057] A distributed sensing unit acquires patient sign physiological parameters in a distributed manner through sensors, the sensors include a heart rate sensor, a blood pressure sensor, an oxygen saturation sensor, a motion sensor and a monitoring camera, and are linked to a hospital information system, traverse the hospital information system, grab preoperative monitoring information, intraoperative monitoring information and postoperative monitoring information from the hospital information system, combine the preoperative monitoring information, the intraoperative monitoring information and the postoperative monitoring information with the patient sign physiological parameters, and obtain the combined patient sign physiological parameters.

[0058] The data interaction unit interacts with a hospital information system containing patient basic information based on Internet of Things technology to obtain the patient basic information.

[0059] The filter denoising unit is configured to load the combined patient physiological parameters, filter and denoise the patient physiological parameters, and intercept the patient physiological parameters based on the number of patient physiological parameter peaks to obtain at least one parameter time window.

[0060] The event extraction unit is configured to extract a parameter time window containing an incidental event based on a probability-driven mechanism, use the parameters in the parameter time window containing the incidental event as incidental physiological parameters, use the mean value of the parameters in the parameter time window not containing the incidental event as a virtual physiological parameter, and integrate the virtual physiological parameters and the incidental physiological parameters as denoised patient physiological parameters.

[0061] The multi-modal fusion unit is configured to obtain the denoised patient physiological parameters and the patient basic information, align the patient physiological parameters and the patient basic information based on a timestamp alignment algorithm, fuse the patient physiological parameters and the patient basic information based on a federated feature fusion method, and output preprocessed patient monitoring data.

[0062] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0063] In the embodiments of the present application, various physiological parameters of a patient are captured and recorded in real time by a sensor technology and a data acquisition device, and at the same time, multi-source information such as the patient's anesthesia medication record, operation details and past medical history is combined to form a complete and comprehensive patient data file. On this basis, a patient evaluation model during the anesthesia recovery period is used to perform multi-dimensional parallel evaluation and analysis on patient monitoring data, and the nursing classification needs of the patient are determined in an intelligent manner. The present application solves the problem of over-reliance on subjective experience of medical staff and limited bedside observation in traditional nursing, ensures the objectivity and scientificity of nursing classification decision-making, helps to optimize the allocation of nursing resources, improves the quality of nursing, reduces the risk of complications, and improves the comfort and satisfaction of patients.

[0064] In the embodiments of the present application, when preprocessing patient monitoring data, the patient physiological parameters are filtered and denoised, and the patient physiological parameters are intercepted based on the number of patient physiological parameter peaks. Different types of parameters can be distinguished and filtered, so that interference noise can be effectively filtered, the data integrity and authenticity are ensured, and the number of peaks and the probability-driven mechanism can be combined to accurately capture incidental physiological events. By dynamically intercepting the parameter time window, only the key event segment is retained, and the amount of data is significantly reduced.

[0065] In the embodiment of the present application, when filtering and denoising the physiological parameters of patient signs, first, the effective signal is dynamically screened through the similarity coefficient, so as to eliminate invalid data such as electrode falling off, and the signal complexity can be effectively quantified by calculating the permutation entropy, and the frequency band is accurately divided, the low-frequency signal is filtered and processed by using the Butterworth low-pass filter, the trend signal is filtered and processed by using the Savitzky-Golay filter, and the high-frequency signal is filtered and processed by using the notch filter, so that different frequency bands are processed differently, the parameter zero distortion is retained, the false judgment caused by noise interference is reduced by removing noise and retaining key characteristics, and the workload of manual data processing of medical staff is reduced by automatic filtering and denoising process, and the work efficiency is improved.

[0066] In the embodiment of the present application, when the parameter time window containing the incidental event is extracted combined with the probability driving mechanism, the TOPSIS model combined with the subjective weight and the objective weight, and the dynamically adjusted parameter variation threshold and the probability integral threshold can accurately capture the incidental event, reduce false positives and false negatives, and the virtual sign physiological parameter technology can quickly compress invalid data, thereby reducing the model data processing load, and only when the abnormality is continuous and cross-parameter coordination, the alarm is triggered, which can solve the dilemma of transient interference and gradual event.

[0067] In the embodiment of the present application, the patient sign physiological parameters and the patient basic information are fused by using the federated feature fusion method, the FedBN adjustment model is used for federated transfer learning, and the encrypted exchange attention matrix is constructed, so that the fusion processing of multi-modal data is realized, not only the patient privacy is effectively protected, but also the data consistency is enhanced, the calculation resource allocation is optimized, the early warning delay is reduced, and high-quality data support is provided for intelligent evaluation of patients in the anesthesia recovery period.

[0068] In the embodiment of the present application, when the anesthesia recovery period patient evaluation model is trained, the modeling sample set is preprocessed by using the hybrid enhancement method, which can effectively increase the diversity and complexity of the data and reduce the risk of overfitting, and the pre-constructed anesthesia recovery period patient evaluation model is iteratively trained by using the multi-stage transfer learning strategy, which can fully utilize the knowledge of the pre-trained model, quickly adapt to new tasks, select sample features higher than the frequency threshold based on the preset frequency threshold, and prune sample features lower than the frequency threshold, thereby significantly reducing the calculation complexity and storage requirement of the model. Through the above training method, the pre-constructed anesthesia recovery period patient evaluation model can output high-quality evaluation results, significantly improve the reliability of clinical decision support, and the lightweight model can quickly run on the edge device to realize real-time evaluation and early warning.

[0069] In the embodiment of the present application, the anesthesia recovery period patient evaluation model takes the convolutional neural network as the initial model, and improves the convolutional neural network, embeds a multi-task output layer between the convolutional neural network and the output layer, and the multi-task output layer is composed of a state evaluation block, a nursing evaluation block and a PPL prediction block, so that the model can output the real-time state evaluation grade, nursing demand grade and anesthesia recovery period patient classification of the patient at the same time, the state evaluation block adopts the pulse neural network, and introduces the decision level fusion algorithm, can simulate the pulse behavior of biological neurons, improve the evaluation accuracy of the real-time state evaluation grade, and improve the multi-dimensional evaluation demand adaptability of the anesthesia recovery period patient. BRIEF DESCRIPTION OF DRAWINGS

[0070] Figure 1 It is the implementation process schematic diagram of the anesthesia recovery period patient classification intelligent evaluation method provided by the present application.

[0071] Figure 2 The implementation process schematic diagram of the patient sign monitoring data preprocessing method is shown.

[0072] Figure 3 The implementation process schematic diagram of the patient sign physiological parameter filtering and denoising processing method is shown.

[0073] Figure 4 The implementation process schematic diagram of the parameter time window extraction method combined with the probability driving mechanism and containing incidental events is shown.

[0074] Figure 5 The implementation process schematic diagram of the method for fusing patient sign physiological parameters and patient basic information by using the federal feature fusion method is shown.

[0075] Figure 6 The implementation process schematic diagram of the method for training the anesthesia recovery period patient evaluation model by using the modeling sample set is shown.

[0076] Figure 7 The implementation process schematic diagram of the multi-dimensional parallel evaluation and analysis method of the anesthesia recovery period patient evaluation model for patient sign monitoring data is shown.

[0077] Figure 8 The structure schematic diagram of the anesthesia recovery period patient classification intelligent evaluation system is shown. DETAILED DESCRIPTION

[0078] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used in the description herein is for describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising," "comprises" and "including" used herein are synonymous with and mean "including, but not limited to"; the terms "first," "second," and the like, as used herein do not have their ordinary meaning, but are used only to distinguish one element from another.

[0079] At present, it is difficult to comprehensively and continuously monitor the subtle changes of a large number of physiological parameters of a patient by using a traditional evaluation method, and the evaluation result has a large subjectivity. In view of the above problems, an intelligent evaluation system and method for grading a patient in a recovery period from anesthesia are provided. In brief, when the method is implemented, first, patient physical sign monitoring data are collected in real time based on a sensor and Internet of Things technology, the patient physical sign monitoring data are preprocessed, a patient in a recovery period from anesthesia evaluation model performs multi-dimensional parallel evaluation and analysis on the patient physical sign monitoring data, and an evaluation and analysis result of the patient is output. Finally, it is determined whether a real-time state evaluation grade exceeds a preset patient risk threshold, a warning signal / continuous monitoring signal, a nursing demand grade, a patient grading in a recovery period from anesthesia are visually presented, and the nursing demand grade and the patient grading in a recovery period from anesthesia are pushed. In the embodiment of the application, through a sensor technology and a data acquisition device, various physiological parameters of a patient are captured and recorded in real time. Meanwhile, in combination with a plurality of source information such as an anesthesia medication record, operation details and a past medical history of the patient, a complete and comprehensive patient data file is formed. On this basis, a patient in a recovery period from anesthesia evaluation model is used to perform multi-dimensional parallel evaluation and analysis on patient physical sign monitoring data, so as to determine a nursing grading demand of the patient in an intelligent manner. The limitations of excessive dependence on subjective experience of medical staff and limited bedside observation in traditional nursing are solved, the objectivity and scientificity of nursing grading decision are ensured, nursing resource allocation is optimized, nursing quality is improved, complication risk is reduced, and patient comfort and satisfaction are improved.

[0080] Embodiment 1

[0081] The embodiment of the application provides an intelligent evaluation method for grading a patient in a recovery period from anesthesia, Figure 1 The intelligent evaluation method for grading a patient in a recovery period from anesthesia is shown in a flowchart, and specifically comprises the following steps.

[0082] S10, patient physical sign monitoring data are collected in real time based on a sensor and Internet of Things technology, the patient physical sign monitoring data are preprocessed, and the preprocessed patient physical sign monitoring data are uploaded to a historical database;

[0083] S20, pre-constructing a patient evaluation model based on a convolutional neural network during anesthesia recovery period, and traversing a historical database to grab a modeling sample set from the historical database, dividing the modeling sample set into an identifiable data format, and training the patient evaluation model during anesthesia recovery period using the modeling sample set;

[0084] S30, obtaining the pre-processed patient sign monitoring data, the patient evaluation model during anesthesia recovery period performing multi-dimensional parallel evaluation and analysis on the patient sign monitoring data, and outputting an evaluation and analysis result of the patient, the evaluation and analysis result including a real-time state evaluation grade of the patient, a nursing requirement grade, and a patient classification during anesthesia recovery period;

[0085] S40, loading the evaluation and analysis result, and judging whether the real-time state evaluation grade exceeds a preset patient risk threshold; it should be noted that the preset patient risk threshold can be 0.6-0.8.

[0086] S50, if the preset patient risk threshold is exceeded, triggering a warning signal; at the same time, when the real-time state evaluation grade exceeds the preset upper limit value 0.8, a first-level warning signal is triggered, and a prompt is given through a sharp alarm sound and a red light, and when the real-time state evaluation grade is between 0.6 and 0.8, a second-level warning signal is triggered, and a prompt is given through a soft alarm sound and a yellow light, so that medical staff can take timely measures.

[0087] S60, if the preset patient risk threshold is not exceeded, triggering a continuous monitoring signal;

[0088] S70, in response to the real-time state evaluation level judgment result, visualizing presenting the early warning signal / continuous monitoring signal, the nursing demand level, the PACU patient level, and pushing the nursing demand level, the PACU patient level, it is to be explained that when the early warning signal / continuous monitoring signal, the nursing demand level, the PACU patient level are visualized presented, the data visualization module presents the patient evaluation analysis result in the form of chart and curve and saves, medical staff can view the historical data of patient at any time, carries out statistical analysis and data mining, so that medical staff can intuitively understand the recovery of patient, wherein, for "the real-time state evaluation level of patient", "the nursing demand level" and "the PACU patient level", it can be understood that the three constitute a progressive clinical decision chain, the real-time state evaluation level of patient is the result obtained after the multi-dimensional analysis of the patient sign monitoring data by the PACU patient evaluation model, it reflects the current physiological state and health condition of patient, is the basic level of evaluation, the real-time state evaluation level is the starting point of the whole clinical decision chain, it carries out statistical analysis to the pulse timestamp, pulse sequence entropy and feature autocorrelation in feature extraction set by pulse neural network, and is output based on decision level fusion algorithm.This level provides basic data support for subsequent nursing demand evaluation, and the nursing demand level is calculated based on the real-time state evaluation level of patient by the CFCs algorithm of nursing evaluation block, it reflects the current nursing demand intensity of patient, is the intermediate level of clinical decision chain, the PACU patient level (PPL) is calculated based on the real-time state evaluation level of patient and the nursing demand level by the L-M algorithm of PPL prediction block, it reflects the overall health condition and nursing demand of patient in PACU, is the final level of clinical decision chain, and the PPL level is the final output of the whole clinical decision chain, it integrates the real-time state evaluation level of patient and the nursing demand level, provides a comprehensive patient level result for medical staff, facilitates the optimization of nursing resource allocation and the development of individualized nursing plan, through this progressive clinical decision chain and multi-dimensional analysis, the application can more comprehensively and scientifically evaluate the nursing demand of patient in PACU, optimize the allocation of nursing resources, improve the quality of nursing, and reduce the risk of complications.

[0089] In the embodiment of the present application, through sensor technology and data acquisition equipment, the physiological parameters of the patient are captured and recorded in real time, at the same time, combined with the multi-source information such as the anesthesia medication record, the operation details and the past medical history of the patient, a complete and comprehensive patient data file is formed, on this basis, the patient sign monitoring data is multi-dimensionally and parallelly evaluated and analyzed by using the anesthesia recovery period patient evaluation model, the nursing grading needs of the patient are determined in an intelligent way, the over-reliance on subjective experience of medical staff and the limitation of limited bed-side observation in traditional nursing are solved, the objectivity and scientificity of the nursing grading decision are ensured, the nursing resource allocation is optimized, the nursing quality is improved, the complication risk is reduced, and the comfort and satisfaction of the patient are improved.

[0090] The embodiment of the present application provides a patient sign monitoring data preprocessing method, Figure 2 The embodiment of the present application provides a patient sign monitoring data preprocessing method,

[0091] S101, patient sign physiological parameters are acquired by sensors in a distributed manner, the sensors include a heart rate sensor, a blood pressure sensor, an oxygen saturation sensor, an end-tidal carbon dioxide sensor, a motion sensor, a monitoring camera and the like, and are linked with a hospital information system, preoperative monitoring information, intraoperative monitoring information and postoperative monitoring information are grabbed from the hospital information system by traversing the hospital information system, and the preoperative monitoring information, the intraoperative monitoring information and the postoperative monitoring information are combined with the patient sign physiological parameters to obtain combined patient sign physiological parameters;

[0092] It should be noted that the patient sign physiological parameters include but are not limited to heart rate, blood pressure, oxygen saturation, respiratory rate, peak airway pressure, pain score, bispectral index, body temperature, blood glucose, urine output, body movement frequency, facial expression, limb convulsion, turning over frequency, postoperative delirium risk score. The preoperative monitoring information includes preoperative baseline indicators: age, vital signs, consciousness, airway, laboratory examination and the like; the intraoperative monitoring information includes operation type, operation time, anesthesia monitoring change, anesthesia method, vital signs and the like; the postoperative monitoring information includes the state of entering the anesthesia recovery room: consciousness, whether there is an airway intubation, ventilation state, vital signs and the like.

[0093] S102, patient basic information is acquired by interacting with a hospital information system containing the patient basic information based on Internet of Things technology; wherein the Internet of Things technology includes but is not limited to LoRaWAN network, 5G / NB-IoT technology, Zigbee 3.0 technology.

[0094] It should be noted that the patient basic information includes but is not limited to patient ID, gender, age, height, weight, past medical history and comorbidities, medication allergy information, operation related information.

[0095] S103, load the merged patient sign physiological parameters, filter and denoise the patient sign physiological parameters, and intercept the patient sign physiological parameters based on the number of patient sign physiological parameter peaks to obtain at least one parameter time window;

[0096] S104, extract the parameter time window containing the incidental event based on the probability driving mechanism, take the parameters in the parameter time window containing the incidental event as the incidental physiological parameters, take the average of the parameters in the parameter time window not containing the incidental event as the virtual sign physiological parameters, integrate the virtual sign physiological parameters and the incidental physiological parameters as the denoised patient sign physiological parameters;

[0097] S105, obtain the denoised patient sign physiological parameters and patient basic information, align the patient sign physiological parameters and the patient basic information based on a timestamp alignment algorithm, fuse the patient sign physiological parameters and the patient basic information by using a federated feature fusion method, and output the preprocessed patient sign monitoring data.

[0098] In the embodiment of the present application, when preprocessing the patient sign monitoring data, the patient sign physiological parameters are filtered and denoised, and the patient sign physiological parameters are intercepted based on the number of patient sign physiological parameter peaks, so that different types of parameters can be distinguished and filtered, thereby effectively filtering interference noise and ensuring data integrity and authenticity. The combination of the peak value number and the probability driving mechanism can accurately capture incidental physiological events, and by dynamically intercepting the parameter time window, only the key event segment is retained, thereby significantly reducing the data volume.

[0099] The embodiment of the present application provides a method for filtering and denoising patient sign physiological parameters, Figure 3 A method for filtering and denoising patient sign physiological parameters is shown, and the method for filtering and denoising patient sign physiological parameters specifically includes:

[0100] S201, similarity detection of patient sign physiological parameters based on template matching technology, patient sign physiological parameters with a similarity coefficient exceeding a similarity threshold are selected by a pre-set similarity threshold, and patient sign physiological parameters exceeding the similarity threshold are regarded as effective patient sign physiological parameters;

[0101] Wherein, the similarity coefficient of the patient sign physiological parameters is calculated by the following formula:

[0102]

[0103] Wherein, A i represents the similarity coefficient of the patient sign physiological parameters, x i , a patient physiological parameter i in a collection period and a mean value of the patient physiological parameter i in a parameter template;

[0104] S202, loading an effective patient physiological parameter, decomposing and reconstructing the patient physiological parameter based on a CEEMD method, dividing the patient physiological parameter into at least one IMF component, and calculating permutation entropy of the patient physiological parameter;

[0105] S203, dividing the IMF component of the patient physiological parameter into a low-frequency signal, a trend signal and a high-frequency signal based on a preset entropy value interval;

[0106] wherein the permutation entropy of the patient physiological parameter is calculated by the following formula:

[0107]

[0108] wherein H p (m, i) is the permutation entropy of the patient physiological parameter, m represents the embedding dimension of the patient physiological parameter, and p(m) represents the parameter permutation mode occurrence probability;

[0109] S204, loading the divided low-frequency signal, trend signal and high-frequency signal, filtering and processing the low-frequency signal by using a Butterworth low-pass filter, filtering and processing the trend signal by using a Savitzky-Golay filter, and filtering and processing the high-frequency signal by using a notch filter;

[0110] wherein when the low-frequency signal is filtered and processed by using the Butterworth low-pass filter, the transfer function is represented as:

[0111]

[0112] wherein H(s) represents the transfer function, s is a complex frequency variable, N represents the order of the Butterworth low-pass filter, p k is an input of the kth pole;

[0113] when the trend signal is filtered and processed by using the Savitzky-Golay filter, the trend signal is extracted by using polynomial fitting, and the polynomial is represented as:

[0114]

[0115] wherein Q i represents a polynomial fitting function, o represents a half window width, c j ,x i+j are least square fitting coefficients, a trend signal fitting input is represented as f s ,f z are a sampling frequency and a cutoff frequency respectively, and d is a polynomial order;

[0116] When the high-frequency signal is filtered by the notch filter, the transfer function of the notch filter is represented as:

[0117]

[0118] Wherein, H(f) represents the transfer function of the notch filter, f s ,f IMF Respectively represent the high-frequency signal sampling frequency and IMF component.

[0119] In the embodiment of the present application, when filtering and denoising the physiological parameters of patient signs, first, the effective signal is dynamically screened through the similarity coefficient, so as to eliminate invalid data such as electrode falling off, and the signal complexity can be effectively quantified by calculating the permutation entropy, and the frequency band is accurately divided. For low-frequency signals, a Butterworth low-pass filter is used for filtering processing, for trend signals, a Savitzky-Golay filter is used for filtering processing, and for high-frequency signals, a notch filter is used for filtering processing, so that different frequency bands are processed differently, the parameter zero distortion is retained, the noise is removed and the key characteristics are retained, the misjudgment caused by noise interference is reduced, and the workload of manual data processing of medical staff is reduced by automatic filtering and denoising process, thereby improving the work efficiency.

[0120] The embodiment of the present application provides a method for extracting a parameter time window containing an occasional event combined with a probability driving mechanism, Figure 4 The method for extracting a parameter time window containing an occasional event combined with a probability driving mechanism is shown in the implementation process schematic diagram, and the method specifically comprises the following steps:

[0121] S301, the subjective weights of different types of parameters are determined based on the expert consultation method, the objective weights of different types of parameters are determined by the entropy weight method, the weight decision matrix containing the subjective weights and the objective weights is constructed combined with the TOPSIS model, and the parameter combination weight is obtained based on the weight decision matrix. In the embodiment, the TOPSIS weight decision matrix is constructed by combining the subjective weight of the expert and the objective weight of the entropy weight method, and the limitation of the traditional method depending on a single weight is solved;

[0122] S302, the parameter variation threshold and the probability integral threshold of the preset probability driving mechanism are based on the parameter combination weight, at least one group of parameter time windows are traversed, the parameter variation coefficient in the parameter time window is calculated based on the probability driving mechanism, wherein the parameter variation threshold is set to 0.8, and the probability integral threshold can be set to 12;

[0123] S303, whether the parameter variation coefficient in the parameter time window exceeds the preset parameter variation threshold is judged;

[0124] S304, if the parameter variation threshold exceeds the preset parameter variation threshold, the probability driving mechanism places the corresponding parameter variation coefficient in the probability integral space;

[0125] S305, if the parameter variation threshold exceeds the preset parameter variation threshold, the parameter variation coefficient is set to 0;

[0126] S306, when the parameter variation coefficient in the probability integral space exceeds the preset probability integral threshold, the probability driving mechanism triggers the extraction mechanism to extract the parameter time window containing the incidental event;

[0127] S307, for the parameter time window not containing the incidental event, the mean value of the parameters in the parameter time window is used as the virtual sign physiological parameter.

[0128] In the embodiment of the application, when the probability driving mechanism extracts the parameter time window containing the incidental event, by combining the TOPSIS model of subjective weight and objective weight, and dynamically adjusting the parameter variation threshold and the probability integral threshold, the incidental event can be accurately captured, the false positives and false negatives can be reduced, the virtual sign physiological parameter technology can quickly compress invalid data, thereby reducing the model data processing load, only when the anomaly is continuous and cross-parameter coordination, the alarm is triggered, and the dilemma of transient interference and gradual event can be solved.

[0129] The embodiment of the application provides a method for fusing patient sign physiological parameters and patient basic information by using a federal feature fusion method, Figure 5 A federal feature fusion method for fusing patient sign physiological parameters and patient basic information is shown, and the method specifically comprises:

[0130] S401, obtain patient sign physiological parameters and patient basic information, and align the parameter types corresponding to the patient ID by using encrypted hash matching, in the embodiment of the application, the patient ID is aligned by using encrypted hash matching, so that the patient privacy information is not leaked in the cross-institution transmission process;

[0131] S402, filter the parameter types based on the parameter combination weight, and perform federal feature alignment processing on the filtered patient sign physiological parameters and patient basic information, wherein the parameter types are filtered based on the parameter combination weight, and the filtered patient sign physiological parameters and patient basic information are processed by federal feature alignment, so as to ensure the consistency of data of different institutions at the feature level;

[0132] In S403, the patient physiological parameter and the patient basic information after the feature alignment are processed by using the FedBN adjustment model for federal transfer learning, and an encrypted exchange attention matrix is constructed, the patient physiological parameter and the patient basic information are fused by using the federal feature fusion method combined with the encrypted exchange attention matrix, and the preprocessed patient sign monitoring data is output, the FedBN (federal batch normalization) adjustment model can effectively process the difference in data distribution of different institutions and improve the generalization ability of the model.

[0133] In the embodiment of the application, the patient physiological parameter and the patient basic information are fused by using the federal feature fusion method, the FedBN adjustment model is used for federal transfer learning, and the encrypted exchange attention matrix is constructed, so that the fusion processing of the multi-modal data is realized, the patient privacy is effectively protected, the data consistency is enhanced, the calculation resource allocation is optimized, the early warning delay is reduced, and high-quality data support is provided for intelligent evaluation of the anesthesia recovery period patient.

[0134] The embodiment of the application provides a method for training an anesthesia recovery period patient evaluation model by using a modeling sample set, Figure 6 A method for training an anesthesia recovery period patient evaluation model by using a modeling sample set is shown, and the method for training an anesthesia recovery period patient evaluation model by using a modeling sample set specifically comprises:

[0135] In S501, the modeling sample set is loaded, the modeling sample set is preprocessed by using a hybrid enhancement method, and the modeling sample set is divided into a training set and a test set, wherein the ratio of the training set to the test set can be 4:1;

[0136] In S502, a pre-constructed anesthesia recovery period patient evaluation model is loaded, a parameter optimizer, a training round, a data batch, a hyperparameter and an early stop condition of the anesthesia recovery period patient evaluation model are preset, wherein the parameter optimizer is an RAdam optimizer, the training round is 200-220, the data batch is 10-18, and the early stop condition is that the joint loss function of the anesthesia recovery period patient evaluation model does not decrease for 4 consecutive rounds.

[0137] In S503, the training set is obtained, the pre-constructed anesthesia recovery period patient evaluation model is iteratively trained by using a multi-stage transfer learning strategy, Fourier transform is performed on the training samples in the training set, Fourier frequency analysis is performed on the sample features after the Fourier transform, sample features higher than a preset frequency threshold are screened based on the frequency threshold, sample features lower than the frequency threshold are pruned, and a lightweight converged anesthesia recovery period patient evaluation model is output.

[0138] In S504, the test set is obtained, the anesthesia recovery period patient evaluation model is executed by taking the test set as input, and a test result is output.

[0139] S505, determining whether the error item of the test result and the true result meets an error threshold, wherein the error threshold can be 0.05-0.1;

[0140] S506, outputting the converged anesthesia recovery period patient evaluation model if the error item of the test result and the true result meets the error threshold;

[0141] If the error item of the test result and the true result does not meet the error threshold, the model hyperparameters are adjusted using the RAdam optimizer, and the pre-constructed anesthesia recovery period patient evaluation model is iteratively trained through the training set combined with the multi-stage transfer learning strategy, returning to step S503.

[0142] In the embodiment of the present application, when training the anesthesia recovery period patient evaluation model, the modeling sample set is preprocessed by the mixed enhancement method, which can effectively increase the diversity and complexity of the data and reduce the risk of overfitting. The pre-constructed anesthesia recovery period patient evaluation model is iteratively trained using the multi-stage transfer learning strategy, which can fully utilize the knowledge of the pre-trained model, quickly adapt to new tasks, filter sample features higher than the frequency threshold based on the preset frequency threshold, and prune sample features lower than the frequency threshold, thereby significantly reducing the computational complexity and storage requirements of the model. Through the above training method, the pre-constructed anesthesia recovery period patient evaluation model can output high-quality evaluation results, significantly improving the reliability of clinical decision support. The lightweight model can quickly run on edge devices to realize real-time evaluation and early warning.

[0143] In the embodiment, the anesthesia recovery patient evaluation model takes a convolutional neural network as an initial model, and introduces an input layer and an output layer in the initial model, the input layer is connected with the convolutional neural network, the convolutional neural network is connected with the output layer, the convolutional neural network includes three convolutional blocks, a maximum pooling layer, and a global pooling layer, the global pooling layer of the convolutional neural network is frozen, a residual connection layer is used to replace the global pooling layer, the residual connection layer is embedded in a residual connection network ResNet, an Embedding processing layer and a 3D Conv layer are introduced between the three convolutional blocks and the maximum pooling layer, the loss function formula of the improved convolutional neural network is a cross-entropy loss function, a multi-task output layer is embedded between the convolutional neural network and the output layer, the multi-task output layer is used for evaluating and outputting the real-time state evaluation level, the nursing demand level, and the anesthesia recovery patient classification of the patient, the multi-task output layer includes a state evaluation block, a nursing evaluation block, and a PPL prediction block, wherein the PPL prediction block (PACU Patient Level, PPL) is used for anesthesia recovery patient classification, the activation function of the state evaluation block is a Softmax function, the activation function of the nursing evaluation block is a Sigmoid function, the activation function of the PPL prediction block is a Linear function, the state evaluation block is a spiking neural network, and the spiking neural network introduces a decision-level fusion algorithm, the nursing evaluation block includes three CFCs layers and a fully connected layer, the PPL prediction block is an LSTM model, and the LSTM model introduces an L-M algorithm.

[0144] The joint loss function of the anesthesia recovery patient evaluation model is represented as:

[0145] L oss =α1L z +α2L h +α3L y (7)

[0146] α1+α2+α3=1(8)

[0147] Wherein, L oss represents the joint loss function, L z , L h , and L y are the loss functions of the state evaluation block, the nursing evaluation block, and the PPL prediction block respectively, and α1, α2, and α3 are the weight coefficients of the state evaluation block, the nursing evaluation block, and the PPL prediction block respectively, in the embodiment, the weight coefficients of the state evaluation block, the nursing evaluation block, and the PPL prediction block can be 0.5, 0.2, and 0.3 respectively, by weighting the joint loss function, the model can dynamically balance the importance of different tasks and ensure that each task can be fully optimized.

[0148] In the embodiment of the present application, the anesthesia recovery patient evaluation model takes a convolutional neural network as an initial model, and improves the convolutional neural network by embedding a multi-task output layer between the convolutional neural network and the output layer, and the multi-task output layer is composed of a state evaluation block, a nursing evaluation block and a PPL prediction block, so that the model can output the real-time state evaluation grade of the patient, the nursing demand grade and the anesthesia recovery patient classification at the same time, the state evaluation block adopts a pulse neural network, and a decision-level fusion algorithm is introduced, which can simulate the pulse behavior of biological neurons, improve the real-time state evaluation grade evaluation accuracy, and improve the multi-dimensional evaluation requirement adaptability of the anesthesia recovery patient.

[0149] The anesthesia recovery patient evaluation model provides a multi-dimensional parallel evaluation and analysis method for patient sign monitoring data, Figure 7 The anesthesia recovery patient evaluation model provides a multi-dimensional parallel evaluation and analysis method for patient sign monitoring data,

[0150] S601, obtain the preprocessed patient sign monitoring data, the input layer of the anesthesia recovery patient evaluation model performs step-up encoding processing on the patient sign monitoring data to obtain a step-up encoded data set, and the input layer performs step-up encoding processing on the patient sign monitoring data, which can unify physiological parameters of different dimensions to a higher dimensional feature space, and enhance the fusion ability of the model to multi-source data;

[0151] S602, load the data step-up set and input the data step-up set into the convolutional neural network, the convolutional neural network identifies the parameter type, and performs convolution fusion processing on the data step-up set based on the parameter type, and outputs a feature extraction set, wherein for electrocardiogram (ECG) signal, CNN can extract key features of arrhythmia; for blood pressure signal, CNN can extract key features of blood pressure fluctuation. Through convolution fusion processing, the model can more accurately identify the physiological state of the patient, and improve the accuracy of feature extraction;

[0152] S603, based on the pulse neural network, the pulse timestamp, the pulse sequence entropy and the feature self-correlation in the feature extraction set are counted, and the feature fusion vector of the feature extraction set is calculated based on the pulse timestamp, the pulse sequence entropy and the feature self-correlation, and the feature fusion vector is weighted and fused based on the decision-level fusion algorithm to output the real-time state evaluation grade of the patient, and based on the pulse neural network (SNN), the pulse timestamp, the pulse sequence entropy and the feature self-correlation in the feature extraction set are counted, which can simulate the pulse behavior of biological neurons, and improve the clinical interpretability of the model. Through the decision-level fusion algorithm, the feature fusion vector is weighted and fused, which can further improve the decision-making ability of the model;

[0153] S604, obtaining the real-time state evaluation level of the patient, and the nursing evaluation block approximates the analytical solution of the patient's nursing demand based on the CFCs algorithm combined with the real-time state evaluation level, and takes the analytical solution of the patient's nursing demand as the nursing demand level of the patient;

[0154] wherein the nursing demand level of the patient is calculated by the following formula:

[0155]

[0156] wherein w h represents the nursing demand level of the patient, w z is the real-time state evaluation level of the patient, x T and l are respectively the feature fusion vector of the feature extraction set, the feature quantity, and H represents the nursing basic level;

[0157] S605, taking the real-time state evaluation level of the patient and the nursing demand level of the patient as the input of the PPL prediction block, constructing the anesthesia recovery period patient grading target function, and the PPL prediction block calculates the anesthesia recovery period patient grading target function based on the L-M algorithm, and outputs the anesthesia recovery period patient grading;

[0158] S606, obtaining the real-time state evaluation level, the nursing demand level and the anesthesia recovery period patient grading of the patient, and integrating the real-time state evaluation level, the nursing demand level and the anesthesia recovery period patient grading of the patient into the evaluation analysis result of the patient.

[0159] Embodiment 2

[0160] The embodiment of the present application also provides a kind of anesthesia recovery period patient grading intelligent evaluation system, Figure 8 It shows the structure schematic diagram of anesthesia recovery period patient grading intelligent evaluation system, the anesthesia recovery period patient grading intelligent evaluation system, specifically includes:

[0161] Data acquisition module 100, based on sensor and Internet of Things technology, real-time acquisition patient sign monitoring data, pre-processes patient sign monitoring data, and uploads the pre-processed patient sign monitoring data to historical database;

[0162] Intelligent evaluation module 200, for constructing anesthesia recovery period patient evaluation model based on convolutional neural network, and obtaining pre-processed patient sign monitoring data, anesthesia recovery period patient evaluation model carries out multi-dimensional parallel evaluation analysis to patient sign monitoring data, and outputs the evaluation analysis result of patient, and evaluation analysis result includes the real-time state evaluation level of patient, nursing demand level, anesthesia recovery period patient grading;

[0163] The early warning alarm module 300 is used for loading the evaluation analysis result, judging whether the real-time state evaluation level exceeds the preset patient risk threshold value, triggering an early warning signal if the preset patient risk threshold value is exceeded, and triggering a continuous monitoring signal if the preset patient risk threshold value is not exceeded.

[0164] The data visualization module 400 is used for responding to the real-time state evaluation level judgment result, visualizing the early warning signal / continuous monitoring signal, the nursing demand level, and the anesthesia recovery period patient classification, and pushing the nursing demand level and the anesthesia recovery period patient classification.

[0165] In the embodiment, the data acquisition module 100 comprises:

[0166] The distributed sensing unit 110 is used for acquiring patient physiological parameters by means of distributed sensors, wherein the sensors comprise a heart rate sensor, a blood pressure sensor, a blood oxygen saturation sensor, a motion sensor, and a monitoring camera, and the distributed sensing unit 110 is linked to a hospital information system, traverses the hospital information system, and extracts preoperative monitoring information, intraoperative monitoring information, and postoperative monitoring information from the hospital information system, and combines the preoperative monitoring information, the intraoperative monitoring information, and the postoperative monitoring information with the patient physiological parameters to obtain combined patient physiological parameters.

[0167] The data interaction unit 120 is used for interacting with a hospital information system containing patient basic information based on Internet of Things technology to obtain the patient basic information.

[0168] The filtering and noise reduction unit 130 is used for loading the combined patient physiological parameters, performing filtering and noise reduction processing on the patient physiological parameters, and performing parameter interception on the patient physiological parameters based on the number of patient physiological parameter peaks to obtain at least one parameter time window.

[0169] The event extraction unit 140 is used for extracting a parameter time window containing an incidental event by means of a probability driving mechanism, taking the parameters in the parameter time window containing the incidental event as incidental physiological parameters, taking the mean value of the parameters in a parameter time window not containing the incidental event as a virtual physiological parameter, integrating the virtual physiological parameters and the incidental physiological parameters as noise-reduced patient physiological parameters, and outputting the noise-reduced patient physiological parameters.

[0170] The multi-modal fusion unit 150 is used for acquiring the noise-reduced patient physiological parameters and the patient basic information, performing alignment processing on the patient physiological parameters and the patient basic information based on a timestamp alignment algorithm, fusing the patient physiological parameters and the patient basic information by means of a federated feature fusion method, and outputting preprocessed patient physiological monitoring data.

[0171] It should be noted that the anesthesia recovery patient grading intelligent evaluation system provided by the embodiments of the present application corresponds to the anesthesia recovery patient grading intelligent evaluation method described above, and the explanation, examples, beneficial effects and the like of the related content can refer to the corresponding content in the anesthesia recovery patient grading intelligent evaluation method, which will not be described here.

[0172] In summary, the present application provides an anesthesia recovery patient grading intelligent evaluation system and method. In the embodiments of the present application, the sensor technology and data acquisition equipment are used to capture and record the physiological parameters of the patient in real time. At the same time, the anesthesia medication records, surgical details, and past medical history of the patient are combined to form a complete and comprehensive patient data file. On this basis, the anesthesia recovery patient evaluation model is used to perform multi-dimensional parallel evaluation and analysis on the patient's physical sign monitoring data, and the patient's nursing grading needs are determined in an intelligent manner. The present application solves the problem of excessive reliance on subjective experience of medical staff and limited bedside observation in traditional nursing, ensures the objectivity and scientificity of nursing grading decision, helps to optimize the allocation of nursing resources, improves the quality of nursing, reduces the risk of complications, and improves the comfort and satisfaction of patients.

[0173] It should be noted that for the foregoing embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited by the order of the described actions, because according to the present application, certain steps can be performed in other order or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0174] The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the protection scope of the application. Obviously, the described embodiments are only some of the embodiments of the present application, not all embodiments. Based on these embodiments, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application. Although the present application has been described in detail with reference to the above embodiments, those of ordinary skill in the art can still combine, delete or make other adjustments to the features in the embodiments of the present application according to the circumstances without making creative labor, so as to obtain different other technical solutions which do not deviate from the concept of the present application in essence. These technical solutions also belong to the scope of the present application.

Claims

1. An intelligent method for grading a patient in a post-anesthesia recovery period, characterized in that, The method comprises: Real-time acquisition of patient sign monitoring data based on sensors and Internet of Things technology, preprocessing of patient sign monitoring data, and uploading of the preprocessed patient sign monitoring data to a historical database; Preconstruction of a patient evaluation model during the anesthesia recovery period based on a convolutional neural network, traversal of the historical database, grabbing of a modeling sample set from the historical database, division of the modeling sample set into an identifiable data format, and training of the patient evaluation model during the anesthesia recovery period using the modeling sample set; Obtaining preprocessed patient sign monitoring data, multi-dimensional parallel evaluation and analysis of the patient sign monitoring data by the patient evaluation model during the anesthesia recovery period, and output of the evaluation and analysis results of the patient, including the real-time state evaluation level, nursing demand level, and patient classification during the anesthesia recovery period; Loading the evaluation and analysis results, determining whether the real-time state evaluation level exceeds a preset patient risk threshold, triggering an early warning signal if the preset patient risk threshold is exceeded, and triggering a continuous monitoring signal if the preset patient risk threshold is not exceeded; In response to the real-time state evaluation level determination result, the early warning signal / continuous monitoring signal, the nursing demand level, and the patient classification during the anesthesia recovery period are visually presented, and the nursing demand level and the patient classification during the anesthesia recovery period are pushed.

2. The method of claim 1, wherein the method further comprises: determining a level of consciousness of the patient; and determining a level of sedation of the patient. The preprocessing method for patient sign monitoring data comprises: Distributed acquisition of patient sign physiological parameters by sensors, including heart rate sensors, blood pressure sensors, blood oxygen saturation sensors, end-tidal carbon dioxide sensors, motion sensors, and monitoring cameras, and linking with a hospital information system, traversing the hospital information system, grabbing preoperative monitoring information, intraoperative monitoring information, and postoperative monitoring information from the hospital information system, merging the preoperative monitoring information, intraoperative monitoring information, and postoperative monitoring information with the patient sign physiological parameters, and obtaining the merged patient sign physiological parameters; Interaction with the hospital information system containing patient basic information based on Internet of Things technology to obtain patient basic information; Loading the merged patient sign physiological parameters, filtering and denoising the patient sign physiological parameters, and cutting the patient sign physiological parameters based on the number of patient sign physiological parameter peaks to obtain at least one parameter time window; Extracting parameter time windows containing incidental events in combination with a probability-driven mechanism, using the parameters in the parameter time windows containing incidental events as incidental physiological parameters, using the mean value of the parameters in the parameter time windows not containing incidental events as virtual sign physiological parameters, integrating the virtual sign physiological parameters and the incidental physiological parameters as denoised patient sign physiological parameters; Obtaining denoised patient sign physiological parameters and patient basic information, aligning the patient sign physiological parameters and the patient basic information based on a timestamp alignment algorithm, fusing the patient sign physiological parameters and the patient basic information using a federated feature fusion method, and outputting preprocessed patient sign monitoring data.

3. The method of claim 2, wherein the method further comprises: determining a level of consciousness of the patient; and determining a level of sedation of the patient. The method for filtering and denoising patient sign physiological parameters comprises: The template matching technology is used for similarity detection of patient physiological parameters, and patient physiological parameters with a similarity coefficient exceeding a preset similarity threshold are screened out as effective patient physiological parameters. The effective patient physiological parameters are loaded, and the CEEMD method is used for decomposition and reconstruction of the patient physiological parameters, so as to divide the patient physiological parameters into at least one IMF component and calculate the permutation entropy of the patient physiological parameters. The IMF components of the patient physiological parameters are divided into low-frequency signals, trend signals and high-frequency signals based on a preset entropy value interval. The divided low-frequency signals, trend signals and high-frequency signals are loaded, and the low-frequency signals are filtered by a Butterworth low-pass filter, the trend signals are filtered by a Savitzky-Golay filter, and the high-frequency signals are filtered by a notch filter.

4. The method of claim 2, wherein the method further comprises: determining a level of consciousness of the patient; and determining a level of sedation of the patient. The parameter time window extraction method containing sporadic events based on the probability-driven mechanism includes: The subjective weights of different types of parameters are determined based on the expert consultation method, the objective weights of different types of parameters are determined by the entropy weight method, the weight decision matrix containing the subjective weights and the objective weights is constructed based on the TOPSIS model, and the parameter combination weight is obtained based on the weight decision matrix; The parameter mutation threshold and the probability integral threshold of the probability-driven mechanism are preset based on the parameter combination weight, at least one parameter time window is traversed, and the parameter variation coefficient in the parameter time window is calculated based on the probability-driven mechanism; It is judged whether the parameter variation coefficient in the parameter time window exceeds the preset parameter variation threshold, if the parameter variation threshold exceeds the preset parameter variation threshold, the corresponding parameter variation coefficient is placed in the probability integral space by the probability-driven mechanism, if the parameter variation threshold exceeds the preset parameter variation threshold, the parameter variation coefficient is set to 0; When the parameter variation coefficient in the probability integral space exceeds the preset probability integral threshold, the probability-driven mechanism triggers the extraction mechanism to extract the parameter time window containing sporadic events; The parameter mean value in the parameter time window is used as a virtual physiological parameter for the parameter time window not containing sporadic events.

5. The intelligent assessment method for grading patients in the anesthesia recovery period as described in claim 4, characterized in that: The method for fusing patient physiological parameters and patient basic information by using the federated feature fusion method includes: The patient physiological parameters and the patient basic information are obtained, and the parameter types corresponding to the patient ID are aligned by using the encrypted hash matching; The parameter types are screened based on the parameter combination weight, and the federated feature alignment processing is performed on the screened patient physiological parameters and patient basic information; The federated transfer learning is performed on the patient physiological parameters and patient basic information after the feature alignment processing by using the FedBN adjustment model, an encrypted exchange attention matrix is constructed, the patient physiological parameters and patient basic information are fused by using the federated feature fusion method combined with the encrypted exchange attention matrix, and the preprocessed patient physiological monitoring data is output.

6. The method of claim 1, wherein: the patient is a post-anesthesia patient; and the patient's condition is assessed in a post-anesthesia care unit (PACU). The method for training the anesthesia recovery patient evaluation model by using the modeling sample set includes: The modeling sample set is loaded, the modeling sample set is preprocessed by using the hybrid enhancement method, and the modeling sample set is divided into a training set and a test set; Load a pre-built anesthesia recovery patient evaluation model, and preset the parameter optimizer, training round, data batch, hyperparameter, and early stopping condition of the anesthesia recovery patient evaluation model; Obtain a training set, and use a multi-stage transfer learning strategy to iteratively train the pre-built anesthesia recovery patient evaluation model; perform Fourier transform on the training samples in the training set, and perform Fourier frequency analysis on the sample features after the Fourier transform; filter sample features higher than a preset frequency threshold based on the frequency threshold; perform pruning processing on sample features lower than the frequency threshold; and output a lightweight converged anesthesia recovery patient evaluation model. Obtain a test set, input the test set, execute the anesthesia recovery patient evaluation model, output a test result, and determine whether the error items of the test result and the true result meet an error threshold. If the error items of the test result and the true result meet the error threshold, output a converged anesthesia recovery patient evaluation model. If the error items of the test result and the true result do not meet the error threshold, adjust the model hyperparameters using an RAdam optimizer, and continue to iteratively train the pre-built anesthesia recovery patient evaluation model using the training set and the multi-stage transfer learning strategy.

7. The intelligent assessment method for grading patients in the anesthesia recovery period as described in claim 6, characterized in that: The anesthesia recovery patient evaluation model uses a convolutional neural network as an initial model, and introduces an input layer and an output layer in the initial model. The input layer is connected to the convolutional neural network, and the convolutional neural network is connected to the output layer. The convolutional neural network includes three convolutional blocks, a max pooling layer, and a global pooling layer. The global pooling layer of the convolutional neural network is frozen, a residual connection layer is used to replace the global pooling layer, the residual connection layer is embedded in a residual connection network ResNet, an Embedding processing layer and a 3D Conv layer are introduced between the three convolutional blocks and the max pooling layer, and the improved convolutional neural network loss function is a cross-entropy loss function. A multi-task output layer is embedded between the convolutional neural network and the output layer. The multi-task output layer is used to evaluate and output the real-time state evaluation level, nursing demand level, and anesthesia recovery patient classification of the patient. The multi-task output layer includes a state evaluation block, a nursing evaluation block, and a PPL prediction block. The activation function of the state evaluation block is a Softmax function, the activation function of the nursing evaluation block is a Sigmoid function, and the activation function of the PPL prediction block is a Linear function. The state evaluation block is a spiking neural network, and the spiking neural network introduces a decision-level fusion algorithm. The nursing evaluation block includes three CFCs layers and a fully connected layer. The PPL prediction block is an LSTM model, and the LSTM model introduces an L-M algorithm.

8. The intelligent assessment method for grading patients in the anesthesia recovery period as described in claim 7, characterized in that: The anesthesia recovery patient evaluation model uses a multi-dimensional parallel evaluation and analysis method for patient sign monitoring data, which includes: Obtain preprocessed patient sign monitoring data, perform step-up encoding processing on the patient sign monitoring data by the input layer of the anesthesia recovery patient evaluation model, and obtain a step-up data set after step-up encoding; Load the step-up data set, input the step-up data set into the convolutional neural network, identify the parameter types, perform convolution fusion processing on the step-up data set based on the parameter types, and output a feature extraction set. The pulse neural network-based statistical feature extraction set extracts centralized pulse timestamps, pulse sequence entropy, and feature autocorrelation, and calculates a feature fusion vector of the feature extraction set based on the pulse timestamps, pulse sequence entropy, and feature autocorrelation. The feature fusion vector is weighted and fused based on a decision-level fusion algorithm, and a real-time state evaluation grade of the patient is output. The real-time state evaluation grade of the patient is obtained, and the nursing evaluation block approximates the analytical solution of the patient's nursing demand based on the CFCs algorithm combined with the real-time state evaluation grade, and takes the analytical solution of the patient's nursing demand as the nursing demand grade of the patient. The real-time state evaluation grade of the patient and the nursing demand grade of the patient are taken as inputs of the PPL prediction block, a patient classification target function in the anesthesia recovery period is constructed, the PPL prediction block calculates the patient classification target function in the anesthesia recovery period based on the L-M algorithm, and outputs the patient classification in the anesthesia recovery period. The real-time state evaluation grade of the patient, the nursing demand grade of the patient, and the patient classification in the anesthesia recovery period are obtained, and the real-time state evaluation grade of the patient, the nursing demand grade of the patient, and the patient classification in the anesthesia recovery period are integrated into the evaluation analysis result of the patient.

9. An intelligent system for assessing the level of recovery of a patient under anesthesia, for implementing the method for assessing the level of recovery of a patient under anesthesia according to any one of claims 1 to 8, characterized in that it comprises: The intelligent evaluation system for patient classification in the anesthesia recovery period comprises: The data acquisition module acquires patient physiological parameter data in real time based on sensors and Internet of Things technology, pre-processes the patient physiological parameter data, and uploads the pre-processed patient physiological parameter data to a historical database. The intelligent evaluation module is configured to construct a patient evaluation model in the anesthesia recovery period based on a convolutional neural network, and to obtain the pre-processed patient physiological parameter data. The patient evaluation model in the anesthesia recovery period performs multi-dimensional parallel evaluation and analysis on the patient physiological parameter data, and outputs an evaluation analysis result of the patient. The evaluation analysis result includes a real-time state evaluation grade of the patient, a nursing demand grade of the patient, and a patient classification in the anesthesia recovery period. The early warning and alarm module is configured to load the evaluation analysis result, determine whether the real-time state evaluation grade exceeds a preset patient risk threshold, trigger an early warning signal if the real-time state evaluation grade exceeds the preset patient risk threshold, and trigger a continuous monitoring signal if the real-time state evaluation grade does not exceed the preset patient risk threshold. The data visualization module is configured to respond to the real-time state evaluation grade determination result, visually present the early warning signal / continuous monitoring signal, the nursing demand grade, and the patient classification in the anesthesia recovery period, and push the nursing demand grade and the patient classification in the anesthesia recovery period.

10. The intelligent patient classification system for anesthesia recovery, as recited in claim 9, wherein: The data acquisition module comprises: The distributed sensing unit acquires patient physiological parameters in a distributed manner through sensors, the sensors include a heart rate sensor, a blood pressure sensor, an oxygen saturation sensor, a motion sensor, and a monitoring camera, and are linked to a hospital information system. The distributed sensing unit traverses the hospital information system, extracts preoperative monitoring information, intraoperative monitoring information, and postoperative monitoring information from the hospital information system, combines the preoperative monitoring information, the intraoperative monitoring information, and the postoperative monitoring information with the patient physiological parameters, and obtains combined patient physiological parameters. The data interaction unit interacts with the hospital information system containing patient basic information based on Internet of Things technology, and obtains patient basic information. The filter denoising unit is configured to load the combined patient physiological parameter, filter and denoise the patient physiological parameter, and cut the patient physiological parameter based on a number of peak values of the patient physiological parameter to obtain at least one parameter time window; The event extraction unit is configured to extract a parameter time window containing an episodic event based on a probability driving mechanism, take parameters in the parameter time window containing the episodic event as episodic physiological parameters, take a mean value of parameters in a parameter time window not containing the episodic event as a virtual physiological parameter, and integrate the virtual physiological parameter and the episodic physiological parameter as denoised patient physiological parameters; The multi-modal fusion unit is configured to obtain the denoised patient physiological parameter and patient basic information, align the patient physiological parameter and the patient basic information based on a timestamp alignment algorithm, fuse the patient physiological parameter and the patient basic information by using a federated feature fusion method, and output preprocessed patient physiological monitoring data.

Images