Respirator parameter autonomous generation and evaluation method, system, equipment and medium

Abstract

The invention relates to the technical field of medical instrument application, in particular to a breathing machine parameter autonomous generation and evaluation method, system, equipment and medium, comprising: setting a basic group, training a basic breathing machine parameter prediction model and a basic breathing machine parameter evaluation model by using standardized breathing data of the basic group, training a corresponding migration respirator parameter prediction model and a migration respirator parameter evaluation model by using the standardized respiration data of each patient group; and inputting the standardized patient state data and the expected patient state data of the target patient into a respirator parameter prediction model of the corresponding patient group to obtain a respirator parameter suggestion value, or inputting the standardized patient state data and the respirator parameter suggestion value into a respirator parameter evaluation model of the corresponding patient group to obtain a respirator parameter evaluation value. And obtaining a standardized patient state data prediction value. According to the method, individualized accurate recommendation and evaluation of the parameters of the breathing machine can be realized, and the problem of model adaptation under the condition of scarcity of specific group data is effectively solved.

Description

Technical Field

[0001] This application relates to the field of medical device application technology, specifically to a method, system, device, and medium for autonomous generation and evaluation of ventilator parameters. Background Technology

[0002] Ventilators are critical life support devices for critically ill patients, and their dynamic parameter adjustment directly affects treatment outcomes. In complex cases such as acute respiratory distress syndrome and chronic obstructive pulmonary disease, precise parameter settings are essential for reducing the risk of complications and improving prognosis.

[0003] Traditional ventilator parameter settings and adjustments rely heavily on the personal experience of medical staff, requiring repeated monitoring of blood gas analysis results for an "adjustment-verification" cycle to gradually approach the target physiological state. To improve the scientific rigor and efficiency of decision-making, some data-driven methods for autonomous generation and evaluation of ventilator parameters have emerged in existing technologies. For example, statistical models or simple machine learning models can be built using historical ventilation data to predict the physiological changes that may result from adjusting a certain parameter, thereby providing quantitative references for clinicians.

[0004] However, these existing data-driven methods for guiding ventilator parameter settings still have limitations: existing methods do not systematically classify and model patient populations, ignoring the physiological and pathological differences between different types of patients, resulting in insufficient model generalization ability and limited prediction accuracy; existing methods have relatively simple fusion processing of respiratory data, failing to fully consider the temporal characteristics and heterogeneity of multimodal respiratory data, and lacking an effective mechanism for unified preprocessing and feature fusion of data from different sources and sampling frequencies, resulting in low data utilization and insufficient feature expression; existing technologies lack effective cross-population knowledge transfer mechanisms, making it impossible to quickly build high-performance prediction models on specific populations with limited data, thus limiting their application in diverse clinical scenarios. Summary of the Invention

[0005] To address the shortcomings of existing data-driven methods for guiding ventilator parameter settings, such as a lack of differentiation among different patient groups, insufficient fusion and mining of multimodal time-series data, and difficulty in effectively modeling data-scarce groups, this application provides a method, system, device, and medium for autonomous generation and evaluation of ventilator parameters. By classifying patient groups and constructing transfer learning models, deeply mining multimodal time-series data using an attention-based LSTM network, and employing a phased training strategy, this approach achieves personalized and accurate recommendation of ventilator parameters and evaluation of the application effects of proposed ventilator parameters. This improves the accuracy of ventilator parameter recommendations and effectively solves the model adaptation problem under data scarcity for specific groups.

[0006] In a first aspect, this application provides a method for autonomously generating and evaluating ventilator parameters, comprising the following steps: S1. Classify patient groups and obtain time-series of multimodal respiratory data from different patient groups during ventilator use. The multimodal respiratory data includes patient status data and ventilator parameters. S2. Preprocess the time series of multimodal respiratory data to obtain a standardized respiratory data time series, including a standardized patient status data time series and a standardized ventilator parameter time series; S3. Set a patient group as the base group, and train a base ventilator parameter prediction model using the standardized respiratory data time series of the base group. The base ventilator parameter prediction model is built based on an LSTM network with an attention mechanism. The input includes the standardized patient state data time series within a specified time window ending at the baseline time point and the standardized patient state data at a specified time step after the baseline time point. The output is the suggested value of the ventilator parameter at the baseline time point. During the training of the basic ventilator parameter prediction model, the actual values ​​of ventilator parameters at the baseline time point are extracted from the standardized respiratory parameter time series as true labels. The loss is calculated by comparing the suggested values ​​of ventilator parameters output by the basic ventilator parameter prediction model with the true labels, and the model parameters are updated using the backpropagation algorithm, thereby establishing a mapping relationship between patient status data and ventilator parameters. The baseline ventilator parameter assessment model is trained using a time series of standardized respiratory data from the baseline population. The baseline ventilator parameter assessment model is built on an LSTM network with an attention mechanism. The input includes a time series of standardized patient status data within a specified time window ending at the baseline time point and the proposed values ​​of ventilator parameters at the baseline time point. The output is the predicted value of standardized patient status data at a specified time step after the baseline time point. During the training of the basic ventilator parameter assessment model, the actual values ​​of standardized patient status data extracted from the standardized respiratory parameter time series at a specified time step after the baseline time point are used as the true labels. The loss is calculated by comparing the predicted values ​​of standardized patient status data output by the basic ventilator parameter assessment model with the true labels, and the model parameters are updated using the backpropagation algorithm, thereby establishing a mapping relationship between patient status data and ventilator parameters. S4. Based on the basic ventilator parameter prediction model, train the corresponding patient group's transfer ventilator parameter prediction model using the time series of standardized respiratory data from each patient group; Based on the basic ventilator parameter assessment model, the transfer ventilator parameter assessment model for each patient group is trained using the time series of standardized respiratory data for each patient group. S5. Obtain the patient status data of the target patient and preprocess it to obtain a standardized time series sequence of patient status data. Determine the patient group to which the target patient belongs, and then perform at least one of the following operations: The standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the expected patient status data at a specified time step after time point t are input into the basic ventilator parameter prediction model or the transfer ventilator parameter prediction model of the corresponding patient group to obtain the suggested values ​​of ventilator parameters at time point t. The standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the proposed values ​​of ventilator parameters at time point t are input into the basic ventilator parameter assessment model or the transfer ventilator parameter assessment model of the corresponding patient group to obtain the expected predicted values ​​of patient status data at a specified time step after time point t.

[0007] It should be further noted that in step S1, the patient group is classified based on at least one of the following factors: race, age group, gender, and disease type.

[0008] It should be further noted that in step S1, the patient status data includes demographic data and physiological and pathological data, among which: Demographic data includes height, weight, and age; Physiological and pathological data include human body pH, PaO2 (partial pressure of oxygen in arterial blood), PaCO2 (partial pressure of carbon dioxide in arterial blood), and body temperature.

[0009] It should be further noted that ventilator parameters include ventilator setting parameters and ventilator monitoring parameters, among which: Ventilator settings include PEEP (positive end-expiratory pressure), Vt (tidal volume), and RR (respiratory rate). Ventilator monitoring parameters include PIP (peak inspiratory pressure), Pplat (plateau pressure), and ventilator flow rate.

[0010] It should be further noted that in step S2, preprocessing includes time series imputation, outlier handling, sampling frequency alignment, feature extraction, and normalization, wherein: For time-series filling, a linear interpolation method is used for continuous monitoring data, and a forward filling method is used for set parameters; Outlier handling includes range filtering of parameters based on clinical guideline thresholds and outlier replacement; Sampling frequency alignment involves resampling data with different sampling frequencies to a uniform time interval; Feature extraction includes time-domain feature extraction and frequency-domain feature extraction; The standardization method adopted is the Z-score standardization method.

[0011] It should be further noted that in step S3, the basic ventilator parameter prediction model includes, in sequence, an input layer, a first LSTM layer, a second LSTM layer, an attention layer, a fully connected layer, a dropout layer, and an output layer, wherein: The input layer is used to receive the time sequence of standardized patient status data within a specified time window ending at the baseline time point and the expected standardized patient status data at a specified time step after the baseline time point; The first LSTM layer contains multiple first LSTM neurons, which extract short-term temporal features from the data received by the input layer and output a hidden state containing sequence information. The second LSTM layer contains multiple second LSTM neurons, which are used to receive the hidden states output by the first LSTM layer, extract long-term temporal dependency features, and output an encoded temporal feature vector. The attention layer receives the temporal feature vector, calculates the weights of the features at each time step, summarizes the weighted features, and outputs a context vector. The fully connected layer receives the context vector, performs feature compression and nonlinear transformation, and outputs fused features. The Dropout layer is used to receive fused features. It suppresses overfitting by randomly disconnecting some neurons during the forward propagation phase and outputs a regularized feature vector. The output layer receives the regularized feature vector and generates suggested values ​​for ventilator parameters at the baseline time point through a linear activation function.

[0012] It should be further noted that the specific training steps for the basic ventilator parameter prediction model in step S3 include: S301. Construct a respiratory dataset, which includes multiple sets of samples. Each set of samples includes a standardized patient status data time series within a specified time window ending at the baseline time point, and standardized patient status data at a specified time step after the baseline time point. The actual values ​​of ventilator parameters at the baseline time point are set as the true labels of the sample. The actual values ​​of ventilator parameters are extracted from the standardized respiratory parameter time series. S302. Divide the respiratory dataset into a training set, a test set, and a validation set; S303. Initialize the parameters of the basic ventilator parameter prediction model; S304. Input the training set into the initialized basic ventilator parameter prediction model and output the suggested values ​​of ventilator parameters; S305. Use a loss function to calculate the loss between the actual values ​​of ventilator parameters and the suggested values ​​of ventilator parameters at the baseline time point for each group of samples, and update the network parameters through the backpropagation algorithm; S306. Use the validation set to adjust the hyperparameters until the basic ventilator parameter prediction model converges. Use the test set to evaluate the performance of the basic ventilator parameter prediction model to obtain the trained basic ventilator parameter prediction model.

[0013] It should be further noted that the loss function of the basic ventilator parameter prediction model is the mean squared error loss function.

[0014] It should be further explained that, in step S4, the specific steps for training the transfer ventilator parameter prediction model for each patient group using the standardized respiratory data time series of each patient group based on the basic ventilator parameter prediction model include: S401. Freeze the first LSTM layer, second LSTM layer, attention layer, fully connected layer, and Dropout layer in the well-trained basic ventilator parameter prediction model; S402. Train the frozen model using a time-series sequence of standardized respiratory data from the target patient population, updating only the parameters of the output layer during training; S403. Unfreeze the attention layer, fully connected layer, and Dropout layer, and fine-tune the unfrozen model using a time series of standardized respiratory data from the target patient group to obtain a well-trained transfer ventilator parameter prediction model for the target patient group.

[0015] It should be further noted that in step S403, the fine-tuning training uses a learning rate of 0.0001 and a sample size of ≤16.

[0016] It should be further noted that in step S403, a learning rate decay strategy is used during the fine-tuning training process.

[0017] It should be further noted that step S6 is also included: after outputting the recommended values ​​of ventilator parameters using the basic ventilator parameter prediction model or the transferred ventilator parameter prediction model, the recommended values ​​of ventilator parameters are checked against the clinical safety threshold. When the recommended values ​​of ventilator parameters exceed the safety threshold range, they are adjusted to the closest safety threshold range.

[0018] It should be further noted that in step S3, the basic ventilator parameter assessment model includes, in sequence, an input fusion layer, a first LSTM layer, a second LSTM layer, an attention layer, a fully connected layer, and an output layer, wherein: The input fusion layer receives the standardized patient status data time series and the proposed values ​​of ventilator parameters within a specified time window ending at the baseline time point. It copies the proposed values ​​of ventilator parameters and concatenates them with the standardized patient status data time series in the feature dimension, and outputs the fused time series feature sequence. The first LSTM layer contains multiple first LSTM neurons, which are used to extract short-term temporal features from the fused temporal feature sequence and output a hidden state containing sequence information. The second LSTM layer contains multiple second LSTM neurons, which are used to receive the hidden states output by the first LSTM layer, extract long-term temporal dependency features, and output an encoded temporal feature vector. The attention layer receives the temporal feature vector, calculates the weights of the features at each time step, summarizes the weighted features, and outputs a context vector. The fully connected layer receives the context vector, performs feature compression and nonlinear transformation, and outputs fused features. The Dropout layer is used to receive fused features. It suppresses overfitting by randomly disconnecting some neurons during the forward propagation phase and outputs a regularized feature vector. The output layer receives the regularized feature vector and generates standardized patient state data predictions at a specified time step after the baseline time point through a linear activation function.

[0019] It should be further noted that the specific training steps for the baseline ventilator parameter assessment model in step S3 include: S311. Construct a second respiratory dataset, which includes multiple sets of samples. Each set of samples includes a standardized patient status data time series within a specified time window ending at the baseline time point, the actual values ​​of ventilator parameters at the baseline time point, and the standardized patient status data actual values ​​at a specified time step after the baseline time point are set as the true labels of the sample. The actual values ​​of ventilator parameters and the standardized patient status data actual values ​​at a specified time step after the baseline time point are extracted from the standardized respiratory parameter time series. S312. Divide the respiratory dataset into a training set, a test set, and a validation set; S313. Initialize the parameters of the baseline ventilator parameter assessment model; S314. Input the training set into the initialized baseline ventilator parameter evaluation model, and output the standardized patient status data prediction values ​​at a specified time step after the baseline time point; S315. Use a loss function to calculate the loss between the actual value and the predicted value of the standardized patient state data at a specified time step after the baseline time point for each group of samples, and update the network parameters through the backpropagation algorithm. S316. Use the validation set to adjust the hyperparameters until the basic ventilator parameter evaluation model converges. Use the test set to evaluate the performance of the basic ventilator parameter evaluation model to obtain the trained basic ventilator parameter evaluation model.

[0020] It should be further noted that the loss function of the basic ventilator parameter assessment model is the mean squared error loss function.

[0021] It should be further explained that, in step S4, the specific steps for training the transfer ventilator parameter assessment model for each patient group using the standardized respiratory data time series of each patient group based on the basic ventilator parameter assessment model include: S411. Freeze the first LSTM layer, second LSTM layer, attention layer, fully connected layer, and Dropout layer in the well-trained basic ventilator parameter evaluation model; S412. Train the frozen model using a time-series sequence of standardized respiratory data from the target patient population, updating only the parameters of the output layer during training; S413. Unfreeze the attention layer, fully connected layer, and Dropout layer, and fine-tune the unfrozen model using a time series of standardized respiratory data from the target patient group to obtain a well-trained transfer ventilator parameter evaluation model for the target patient group.

[0022] It should be further noted that in step S413, the fine-tuning training uses a learning rate of 0.0001 and a sample size of ≤16.

[0023] It should be further noted that in step S413, a learning rate decay strategy is used during the fine-tuning training process.

[0024] Secondly, this application provides a system for autonomous generation and evaluation of ventilator parameters, used to implement the aforementioned method for autonomous generation and evaluation of ventilator parameters, including: The multimodal respiratory data acquisition module is used to acquire time-series sequences of multimodal respiratory data from different patient groups; The respiratory data preprocessing module is used to preprocess the time series of multimodal respiratory data to obtain a standardized respiratory data time series. The basic prediction model training module is used to set a patient population as the basic population and train a basic ventilator parameter prediction model using the time series of standardized respiratory data of the basic population. The transfer prediction model training module is used to train the transfer ventilator parameter prediction model for each patient group based on the basic ventilator parameter prediction model and using the time series of standardized respiratory data for each patient group. The suggested value generation module is used to input the standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the expected patient status data at a specified time step after time point t into the basic ventilator parameter prediction model or the transfer ventilator parameter prediction model of the corresponding patient group to obtain the suggested ventilator parameter values ​​at time point t. The prediction value generation module is used to input the standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the proposed values ​​of ventilator parameters at time point t into the basic ventilator parameter assessment model or the transferred ventilator parameter assessment model of the corresponding patient group to obtain the expected predicted values ​​of patient status data at a specified time step after time point t.

[0025] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for autonomous generation and evaluation of ventilator parameters.

[0026] Fourthly, this application provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for autonomously generating and evaluating ventilator parameters.

[0027] As can be seen from the above technical solutions, this application has the following advantages: 1. This application effectively overcomes the shortcomings of existing technologies that ignore group differences by classifying patient groups and constructing basic ventilator parameter prediction models and corresponding transfer ventilator parameter prediction models, basic ventilator parameter evaluation models and corresponding transfer ventilator parameter evaluation models. It achieves targeted adaptation of model parameters through transfer learning mechanism to address the differences in physiological characteristics of different groups, and significantly improves the prediction accuracy and individualized adjustment capability for specific patient groups.

[0028] 2. This application addresses the problem of insufficient temporal feature extraction in existing technologies by introducing an LSTM network with an attention mechanism to perform deep modeling of the temporal sequence of multimodal respiratory data. It adaptively captures long-term dependencies and key time-step features in respiratory data, enabling accurate prediction of ventilator parameter trends and assessment of the impact of proposed ventilator parameters on patient status. This achieves bidirectional temporal relationship mining, improving the reliability and timeliness of ventilator parameter optimization. 3. This application adopts a phased transfer learning training strategy. First, it trains a general feature extraction layer using basic population data. Then, it gradually fine-tunes the model parameters by combining freezing and thawing. This can quickly build a high-performance prediction model under the condition of limited target population data, effectively solving the problem of difficult training of data-scarce population models and improving the practical application value of the model.

[0029] 4. This application combines a ventilator parameter prediction model with an evaluation model to form a two-way mechanism of "parameter recommendation - effect prediction", providing more comprehensive and safer decision support for clinical practice and realizing individualized and closed-loop optimization adjustment of ventilator parameters. Attached Figure Description

[0030] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a flowchart of a method for autonomous generation and evaluation of ventilator parameters in one embodiment of this application.

[0032] Figure 2 This is a schematic block diagram of a system for autonomous generation and evaluation of ventilator parameters in one embodiment of this application.

[0033] Figure 3 This is a schematic diagram of the hardware structure of an electronic device in one embodiment of this application. Detailed Implementation

[0034] To make the purpose, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solution protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] The following describes in detail the method for autonomous generation and evaluation of ventilator parameters involved in this application. Specific details, such as particular system structures and technologies, are presented for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details.

[0036] In the autonomous generation and evaluation method for ventilator parameters involved in this application, the term "comprising" indicates the presence of the described feature, whole, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or sets thereof. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0037] To facilitate a clear description of the technical solutions of this application, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" do not necessarily imply that they are different.

[0038] The terms "one embodiment" or "some embodiments" used in this application mean that one or more embodiments of this application include the specific features, structures, or characteristics described in that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this application do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0039] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0040] The autonomous generation and evaluation method for ventilator parameters provided in this application embodiment is executed by a computer device, and correspondingly, the autonomous generation and evaluation system for ventilator parameters runs in the computer device.

[0041] Figure 1 This is a flowchart of a method for autonomously generating and evaluating ventilator parameters according to an embodiment of this application. Figure 1 The implementing entity can be a system for autonomously generating and evaluating ventilator parameters. Depending on different needs, the order of steps in this flowchart can be changed, and some steps can be omitted.

[0042] like Figure 1 As shown, the method for autonomous generation and evaluation of ventilator parameters includes: Step S1: Classify the patient groups and obtain time-series of multimodal respiratory data from different patient groups during ventilator use. The multimodal respiratory data includes patient status data and ventilator parameters.

[0043] By classifying patients and obtaining their multimodal respiratory time-series data during ventilator use, a systematic classification data foundation is provided for the subsequent development of population-specific predictive models.

[0044] In some specific embodiments, patient groups are classified based on at least one of the following factors: race, age group, gender, and disease type.

[0045] By clearly classifying patient groups based on clinically relevant factors such as race, age group, gender, and disease type, the scientific and rational nature of the classification method is ensured, providing a basis for adapting the model to groups with different physiological characteristics.

[0046] In some specific embodiments, patient status data includes demographic data and physiological and pathological data, wherein: Demographic data includes height, weight, and age; Physiological and pathological data include human body pH, PaO2 (partial pressure of oxygen in arterial blood), PaCO2 (partial pressure of carbon dioxide in arterial blood), and body temperature.

[0047] By clearly defining the specific types of demographic and physiological pathological data included in patient status data, the model is provided with comprehensive and structured input features, enhancing the correlation between parameter recommendations and the actual physiological state of patients.

[0048] In some specific embodiments, in step S1, the ventilator parameters include ventilator setting parameters and ventilator monitoring parameters, wherein: Ventilator settings include PEEP (positive end-expiratory pressure), Vt (tidal volume), and RR (respiratory rate). Ventilator monitoring parameters include PIP (peak inspiratory pressure), Pplat (plateau pressure), and ventilator flow rate.

[0049] Step S2: Preprocess the time series of multimodal respiratory data to obtain a standardized respiratory data time series, including a standardized patient status data time series and a standardized ventilator parameter time series.

[0050] By preprocessing the raw multimodal respiratory time-series data to obtain high-quality, standardized input data, the efficiency of model training and the reliability of prediction results can be significantly improved. In some specific embodiments, preprocessing includes time series imputation, outlier handling, sampling frequency alignment, feature extraction, and normalization, wherein: For time-series filling, a linear interpolation method is used for continuous monitoring data, and a forward filling method is used for set parameters; Outlier handling includes range filtering of parameters based on clinical guideline thresholds and outlier replacement; Sampling frequency alignment involves resampling data with different sampling frequencies to a uniform time interval; Feature extraction includes time-domain feature extraction and frequency-domain feature extraction; The standardization method adopted is the Z-score standardization method.

[0051] By specifying specific preprocessing steps and methods, including time series imputation, outlier handling, frequency alignment, feature extraction, and standardization, the consistency and usability of multi-source heterogeneous time series data after cleaning and transformation are ensured.

[0052] Step S3: Set a patient group as the base group, and train the base ventilator parameter prediction model using the standardized respiratory data time series sequence of the base group. The base ventilator parameter prediction model is built based on an LSTM network with an attention mechanism. The input includes the standardized patient state data time series sequence within a specified time window ending at the baseline time point and the standardized patient state data at a specified time step after the baseline time point. The output is the suggested value of the ventilator parameter at the baseline time point. During the training of the basic ventilator parameter prediction model, the actual values ​​of ventilator parameters at the baseline time point are extracted from the standardized respiratory parameter time series as true labels. The loss is calculated by comparing the suggested values ​​of ventilator parameters output by the basic ventilator parameter prediction model with the true labels, and the model parameters are updated using the backpropagation algorithm, thereby establishing a mapping relationship between patient status data and ventilator parameters. The baseline ventilator parameter assessment model is trained using a time series of standardized respiratory data from the baseline population. The baseline ventilator parameter assessment model is built on an LSTM network with an attention mechanism. The input includes a time series of standardized patient status data within a specified time window ending at the baseline time point and the proposed values ​​of ventilator parameters at the baseline time point. The output is the predicted value of standardized patient status data at a specified time step after the baseline time point. During the training of the basic ventilator parameter assessment model, the actual values ​​of standardized patient status data extracted from the standardized respiratory parameter time series at a specified time step after the baseline time point are used as the true labels. The loss is calculated by comparing the predicted values ​​of standardized patient status data output by the basic ventilator parameter assessment model with the true labels, and the model parameters are updated using the backpropagation algorithm, thereby establishing a mapping relationship between patient status data and ventilator parameters.

[0053] By constructing a basic ventilator parameter prediction model using an attention-based LSTM network structure and training the model with basic population data, the complex temporal dependencies between historical and future patient states can be effectively captured, thereby improving the performance of the basic ventilator parameter prediction model.

[0054] In some specific embodiments, the basic ventilator parameter prediction model includes an input layer, a first LSTM layer, a second LSTM layer, an attention layer, a fully connected layer, a dropout layer, and an output layer connected in sequence, wherein: The input layer is used to receive the time sequence of standardized patient status data within a specified time window ending at the baseline time point and the expected standardized patient status data at a specified time step after the baseline time point; The first LSTM layer contains multiple first LSTM neurons, which extract short-term temporal features from the data received by the input layer and output a hidden state containing sequence information. The second LSTM layer contains multiple second LSTM neurons, which are used to receive the hidden states output by the first LSTM layer, extract long-term temporal dependency features, and output an encoded temporal feature vector. The attention layer receives the temporal feature vector, calculates the weights of the features at each time step, summarizes the weighted features, and outputs a context vector. The fully connected layer receives the context vector, performs feature compression and nonlinear transformation, and outputs fused features. The Dropout layer is used to receive fused features. It suppresses overfitting by randomly disconnecting some neurons during the forward propagation phase and outputs a regularized feature vector. The output layer receives the regularized feature vector and generates suggested values ​​for ventilator parameters at the baseline time point through a linear activation function.

[0055] The following recursive equation illustrates how LSTM layers work.

[0056]

[0057]

[0058]

[0059]

[0060]

[0061] In this process, each unit in the sequence input to the LSTM layer is given an activation state. and cell state and the corresponding input ; , and It is an LSTM gate used for cell state input, output, and forgetting information; and These are the cell memory state vector and the hidden state vector, respectively. Represents a Sigmoid type function; It is the input vector; It is a linear transformation matrix, the parameters of which need to be learned for each gate and unit memory, and It is the corresponding deviation vector; For simplicity, LSTM can be represented in the following form:

[0062] To address the individual-specificity of the disease, an attention module was incorporated into the basic ventilator parameter prediction model. An attention-based LSTM was used to enhance the model's ability to supplement key temporal features and multimodal information. The main weight parameters of the attention mechanism are shown in equation (7).

[0063] in, and These are the weights used when the multilayer perceptron calculates attention. Bias when calculating attention weights for a multilayer perceptron. Is it a hidden layer? Output at any given moment.

[0064] The model employs a self-attention mechanism, which calculates the output sequence of the LSTM. Generate attention weights as shown in equation (8). ,

[0065] in, Is Weighted scores corresponding to different features at different times.

[0066] Attention output compresses temporal features into a fixed-dimensional vector. As shown in equation (9)

[0067] By employing a specific network architecture that includes dual LSTM layers, attention layers, and dropout layers, the model is able to extract short-term and long-term time-series features in a hierarchical manner and focus on key information, while suppressing overfitting and improving the model's expressive power and robustness.

[0068] In some specific embodiments, the specific training steps of the baseline ventilator parameter prediction model include: S301. Construct a respiratory dataset, which includes multiple sets of samples. Each set of samples includes a standardized patient status data time series within a specified time window ending at the baseline time point, and standardized patient status data at a specified time step after the baseline time point. The actual values ​​of ventilator parameters at the baseline time point are set as the true labels of the sample. The actual values ​​of ventilator parameters are extracted from the standardized respiratory parameter time series. S302. Divide the respiratory dataset into a training set, a test set, and a validation set; S303. Initialize the parameters of the basic ventilator parameter prediction model; S304. Input the training set into the initialized basic ventilator parameter prediction model and output the suggested values ​​of ventilator parameters; S305. Use a loss function to calculate the loss between the actual values ​​of ventilator parameters and the suggested values ​​of ventilator parameters at the baseline time point for each group of samples, and update the network parameters through the backpropagation algorithm; S306. Use the validation set to adjust the hyperparameters until the basic ventilator parameter prediction model converges. Use the test set to evaluate the performance of the basic ventilator parameter prediction model to obtain the trained basic ventilator parameter prediction model.

[0069] By clearly defining the complete steps in basic model training, including dataset construction, partitioning, parameter initialization, forward propagation, loss calculation, back propagation, and hyperparameter tuning, the standardization of the model training process and the reproducibility of the results are ensured.

[0070] In some specific embodiments, the loss function of the basic ventilator parameter prediction model is the mean squared error loss function.

[0071] The formula for calculating the mean squared error loss function is shown in equation (10).

[0072] in, For the true value, For predicted values, N This represents the number of samples.

[0073] By using the mean squared error loss function to guide model training, the model parameter optimization process is directly aimed at reducing the error between the predicted value and the true value, thereby improving the accuracy of regression prediction.

[0074] In some specific embodiments, the basic ventilator parameter assessment model includes an input fusion layer, a first LSTM layer, a second LSTM layer, an attention layer, a fully connected layer, and an output layer connected in sequence, wherein: The input fusion layer receives the standardized patient status data time series and the proposed values ​​of ventilator parameters within a specified time window ending at the baseline time point. It copies the proposed values ​​of ventilator parameters and concatenates them with the standardized patient status data time series in the feature dimension, and outputs the fused time series feature sequence. The first LSTM layer contains multiple first LSTM neurons, which are used to extract short-term temporal features from the fused temporal feature sequence and output a hidden state containing sequence information. The second LSTM layer contains multiple second LSTM neurons, which are used to receive the hidden states output by the first LSTM layer, extract long-term temporal dependency features, and output an encoded temporal feature vector. The attention layer receives the temporal feature vector, calculates the weights of the features at each time step, summarizes the weighted features, and outputs a context vector. The fully connected layer receives the context vector, performs feature compression and nonlinear transformation, and outputs fused features. The Dropout layer is used to receive fused features. It suppresses overfitting by randomly disconnecting some neurons during the forward propagation phase and outputs a regularized feature vector. The output layer receives the regularized feature vector and generates standardized patient state data predictions at a specified time step after the baseline time point through a linear activation function.

[0075] In some specific embodiments, step S3, the specific training steps of the baseline ventilator parameter evaluation model include: S311. Construct a second respiratory dataset, which includes multiple sets of samples. Each set of samples includes a standardized patient status data time series within a specified time window ending at the baseline time point, the actual values ​​of ventilator parameters at the baseline time point, and the standardized patient status data actual values ​​at a specified time step after the baseline time point are set as the true labels of the sample. The actual values ​​of ventilator parameters and the standardized patient status data actual values ​​at a specified time step after the baseline time point are extracted from the standardized respiratory parameter time series. S312. Divide the respiratory dataset into a training set, a test set, and a validation set; S313. Initialize the parameters of the baseline ventilator parameter assessment model; S314. Input the training set into the initialized baseline ventilator parameter evaluation model, and output the standardized patient status data prediction values ​​at a specified time step after the baseline time point; S315. Use a loss function to calculate the loss between the actual value and the predicted value of the standardized patient state data at a specified time step after the baseline time point for each group of samples, and update the network parameters through the backpropagation algorithm. S316. Use the validation set to adjust the hyperparameters until the basic ventilator parameter evaluation model converges. Use the test set to evaluate the performance of the basic ventilator parameter evaluation model to obtain the trained basic ventilator parameter evaluation model.

[0076] In some specific embodiments, the loss function of the baseline ventilator parameter assessment model is the mean squared error loss function.

[0077] Step S4: Based on the basic ventilator parameter prediction model, train the transfer ventilator parameter prediction model for the corresponding patient group using the time series of standardized respiratory data for each patient group. Based on the basic ventilator parameter assessment model, a transfer ventilator parameter assessment model for the corresponding patient group is trained using time-series standardized respiratory data of each patient group.

[0078] By starting with a pre-trained basic ventilator parameter prediction model and further training the transfer model using data from specific patient groups, efficient transfer and adaptation of general respiratory mechanics features to different target groups was achieved.

[0079] In some specific embodiments, the specific steps for training a transfer ventilator parameter prediction model for a corresponding patient group using standardized respiratory data time-series sequences based on the baseline ventilator parameter prediction model include: S401. Freeze the first LSTM layer, second LSTM layer, attention layer, fully connected layer, and Dropout layer in the well-trained basic ventilator parameter prediction model; S402. Train the frozen model using a time-series sequence of standardized respiratory data from the target patient population, updating only the parameters of the output layer during training; S403. Unfreeze the attention layer, fully connected layer, and Dropout layer, and fine-tune the unfrozen model using a time series of standardized respiratory data from the target patient group to obtain a well-trained transfer ventilator parameter prediction model for the target patient group.

[0080] By adopting a phased transfer strategy that first freezes the feature layer to train the output layer and then unfreezes some layers for fine-tuning, model adaptation can be completed efficiently and stably even with limited target group data.

[0081] In some specific embodiments, in step S403, the fine-tuning training uses a learning rate of 0.0001 and a sample size of ≤16.

[0082] By using a lower learning rate and smaller batches of samples for training during the fine-tuning phase, the stability and refinement of network parameter updates are ensured, which is beneficial for the model to achieve good generalization performance on limited data.

[0083] In some specific embodiments, in step S403, a learning rate decay strategy is used during the fine-tuning training process.

[0084] By employing a learning rate decay strategy during fine-tuning, the model can update parameters with smaller steps in the later stages of training, which is beneficial for converging to a better performance point and further improving prediction accuracy.

[0085] In some specific embodiments, the specific steps for training a transfer ventilator parameter assessment model for a corresponding patient group using standardized respiratory data time-series sequences based on the baseline ventilator parameter assessment model include: S411. Freeze the first LSTM layer, second LSTM layer, attention layer, fully connected layer, and Dropout layer in the well-trained basic ventilator parameter evaluation model; S412. Train the frozen model using a time-series sequence of standardized respiratory data from the target patient population, updating only the parameters of the output layer during training; S413. Unfreeze the attention layer, fully connected layer, and Dropout layer, and fine-tune the unfrozen model using a time series of standardized respiratory data from the target patient group to obtain a well-trained transfer ventilator parameter evaluation model for the target patient group.

[0086] In some specific embodiments, in step S413, the fine-tuning training uses a learning rate of 0.0001 and a sample size of ≤16.

[0087] In some specific embodiments, in step S413, a learning rate decay strategy is used during the fine-tuning training process.

[0088] Step S5: Obtain the patient status data of the target patient and preprocess it to obtain a standardized patient status data time series. Determine the patient group to which the target patient belongs. Input the standardized patient status data time series of the target patient within a specified time window ending at time point t and the expected patient status data at a specified time step after time point t into the basic ventilator parameter prediction model or the transfer ventilator parameter prediction model of the corresponding patient group to obtain the suggested ventilator parameter values ​​at time point t.

[0089] By selecting the appropriate prediction model based on the target patient's group and inputting their current standardized status data and expected status data, personalized and precise ventilator parameter recommendations for the patient at time point t can be output in real time.

[0090] In some specific embodiments, step S6 is also included: performing clinical safety threshold verification on the recommended values ​​of ventilator parameters, and adjusting them to the closest safe threshold range when the recommended values ​​exceed the safe threshold range.

[0091] By verifying and adjusting the recommended parameter values ​​output by the model with clinical safety thresholds, it was ensured that all recommended parameters were within safe limits, providing ultimate assurance for the reliability of clinical applications and patient safety.

[0092] In one specific embodiment, the steps of the method for autonomous generation and evaluation of ventilator parameters include: Step S1: Classify the patient groups and obtain time-series of multimodal respiratory data from different patient groups during ventilator use. The multimodal respiratory data includes patient status data and ventilator parameters. Among them, patient groups are classified based on their race, age group, gender, and disease type; Patient status data includes demographic data and physiological and pathological data, among which: Demographic data includes height, weight, and age; Physiological and pathological data include human body pH, PaO2 (partial pressure of oxygen in arterial blood), PaCO2 (partial pressure of carbon dioxide in arterial blood), and body temperature; Ventilator parameters include ventilator settings and ventilator monitoring parameters, among which: Ventilator settings include PEEP (positive end-expiratory pressure), Vt (tidal volume), and RR (respiratory rate). Ventilator monitoring parameters include PIP (peak inspiratory pressure), Pplat (plateau pressure), and ventilator flow rate.

[0093] Step S2 involves preprocessing the time series of multimodal respiratory data to obtain a standardized respiratory data time series, including a standardized patient status data time series and a standardized ventilator parameter time series. Preprocessing includes time series imputation, outlier handling, sampling frequency alignment, feature extraction, and standardization. For time-series filling, a linear interpolation method is used for continuous monitoring data, and a forward filling method is used for set parameters; Outlier handling includes range filtering of parameters based on clinical guideline thresholds and outlier replacement; Sampling frequency alignment involves resampling data with different sampling frequencies to a uniform time interval; Feature extraction includes time-domain feature extraction and frequency-domain feature extraction; The standardization method adopted is the Z-score standardization method.

[0094] Step S3: A Caucasian patient population aged 18 to 88 years with COPD (chronic obstructive pulmonary disease) and hospitalized for 24 hours or more, excluding those with severe pneumothorax, pulmonary embolism, or other diseases that interfere with respiratory mechanics assessment, is set as the baseline population. The baseline ventilator parameter prediction model is trained using the standardized respiratory data time series of the baseline population. The baseline ventilator parameter prediction model is built based on an LSTM network with an attention mechanism. The input includes the standardized patient status data time series within a specified time window ending at the baseline time point and the standardized patient status data at a specified time step after the baseline time point. The output is the suggested values ​​of ventilator parameters at the baseline time point. During the training of the basic ventilator parameter prediction model, the actual values ​​of ventilator parameters at the baseline time point are extracted from the standardized respiratory parameter time series as true labels. The loss is calculated by comparing the suggested values ​​of ventilator parameters output by the basic ventilator parameter prediction model with the true labels, and the model parameters are updated using the backpropagation algorithm, thereby establishing a mapping relationship between patient status data and ventilator parameters. The basic ventilator parameter prediction model consists of an input layer, a first LSTM layer, a second LSTM layer, an attention layer, a fully connected layer, a dropout layer, and an output layer connected in sequence, wherein: The input layer is used to receive the time sequence of standardized patient status data within a specified time window ending at the baseline time point and the expected standardized patient status data at a specified time step after the baseline time point; The first LSTM layer contains multiple first LSTM neurons, which extract short-term temporal features from the data received by the input layer and output a hidden state containing sequence information. The second LSTM layer contains multiple second LSTM neurons, which are used to receive the hidden states output by the first LSTM layer, extract long-term temporal dependency features, and output an encoded temporal feature vector. The attention layer receives the temporal feature vector, calculates the weights of the features at each time step, summarizes the weighted features, and outputs a context vector. The fully connected layer receives the context vector, performs feature compression and nonlinear transformation, and outputs fused features. The Dropout layer is used to receive fused features. It suppresses overfitting by randomly disconnecting some neurons during the forward propagation phase and outputs a regularized feature vector. The output layer receives the regularized feature vector and generates suggested values ​​for ventilator parameters at the baseline time point through a linear activation function. The specific training steps for the basic ventilator parameter prediction model include: S301. Construct a respiratory dataset, which includes multiple sets of samples. Each set of samples includes a standardized patient status data time series within a specified time window ending at the baseline time point, and standardized patient status data at a specified time step after the baseline time point. The actual values ​​of ventilator parameters at the baseline time point are set as the true labels of the sample. The actual values ​​of ventilator parameters are extracted from the standardized respiratory parameter time series. S302. Divide the respiratory dataset into a training set, a test set, and a validation set; S303. Initialize the parameters of the basic ventilator parameter prediction model; S304. Input the training set into the initialized basic ventilator parameter prediction model and output the suggested values ​​of ventilator parameters; S305. Use a loss function to calculate the loss between the actual values ​​of ventilator parameters and the suggested values ​​of ventilator parameters at the baseline time point for each group of samples, and update the network parameters through the backpropagation algorithm; S306. Use the validation set to adjust the hyperparameters until the basic ventilator parameter prediction model converges, and use the test set to evaluate the performance of the basic ventilator parameter prediction model to obtain the trained basic ventilator parameter prediction model. The baseline ventilator parameter assessment model is trained using a time series of standardized respiratory data from the baseline population. The baseline ventilator parameter assessment model is built on an LSTM network with an attention mechanism. The input includes a time series of standardized patient status data within a specified time window ending at the baseline time point and the proposed values ​​of ventilator parameters at the baseline time point. The output is the predicted value of standardized patient status data at a specified time step after the baseline time point. During the training of the basic ventilator parameter assessment model, the actual values ​​of standardized patient status data extracted from the standardized respiratory parameter time series at a specified time step after the baseline time point are used as the true labels. The loss is calculated by comparing the predicted values ​​of standardized patient status data output by the basic ventilator parameter assessment model with the true labels, and the model parameters are updated using the backpropagation algorithm, thereby establishing a mapping relationship between patient status data and ventilator parameters. The basic ventilator parameter assessment model consists of an input fusion layer, a first LSTM layer, a second LSTM layer, an attention layer, a fully connected layer, and an output layer connected in sequence, wherein: The input fusion layer receives the standardized patient status data time series and the proposed values ​​of ventilator parameters within a specified time window ending at the baseline time point. It copies the proposed values ​​of ventilator parameters and concatenates them with the standardized patient status data time series in the feature dimension, and outputs the fused time series feature sequence. The first LSTM layer contains multiple first LSTM neurons, which are used to extract short-term temporal features from the fused temporal feature sequence and output a hidden state containing sequence information. The second LSTM layer contains multiple second LSTM neurons, which are used to receive the hidden states output by the first LSTM layer, extract long-term temporal dependency features, and output an encoded temporal feature vector. The attention layer receives the temporal feature vector, calculates the weights of the features at each time step, summarizes the weighted features, and outputs a context vector. The fully connected layer receives the context vector, performs feature compression and nonlinear transformation, and outputs fused features. The Dropout layer is used to receive fused features. It suppresses overfitting by randomly disconnecting some neurons during the forward propagation phase and outputs a regularized feature vector. The output layer receives the regularized feature vector and generates standardized patient state data predictions at a specified time step after the baseline time point through a linear activation function. The specific training steps for the basic ventilator parameter assessment model include: S311. Construct a second respiratory dataset, which includes multiple sets of samples. Each set of samples includes a standardized patient status data time series within a specified time window ending at the baseline time point, the actual values ​​of ventilator parameters at the baseline time point, and the standardized patient status data actual values ​​at a specified time step after the baseline time point are set as the true labels of the sample. The actual values ​​of ventilator parameters and the standardized patient status data actual values ​​at a specified time step after the baseline time point are extracted from the standardized respiratory parameter time series. S312. Divide the respiratory dataset into a training set, a test set, and a validation set; S313. Initialize the parameters of the baseline ventilator parameter assessment model; S314. Input the training set into the initialized baseline ventilator parameter evaluation model, and output the standardized patient status data prediction values ​​at a specified time step after the baseline time point; S315. Use the mean squared error loss function to calculate the loss between the actual value and the predicted value of the standardized patient state data at a specified time step after the baseline time point for each group of samples, and update the network parameters through the backpropagation algorithm. S316. Use the validation set to adjust the hyperparameters until the basic ventilator parameter evaluation model converges. Use the test set to evaluate the performance of the basic ventilator parameter evaluation model to obtain the trained basic ventilator parameter evaluation model.

[0095] Step S4: Based on the basic ventilator parameter prediction model, train the corresponding transfer ventilator parameter prediction model for each patient group using the time-series of standardized respiratory data. Specific steps include: S401. Freeze the first LSTM layer, second LSTM layer, attention layer, fully connected layer, and Dropout layer in the well-trained basic ventilator parameter prediction model; S402. Train the frozen model using a time-series sequence of standardized respiratory data from the target patient population, updating only the parameters of the output layer during training; S403. Unfreeze the attention layer, fully connected layer, and Dropout layer, and fine-tune the unfrozen model using a time series of standardized respiratory data from the target patient group. The fine-tuning training uses a learning rate of 0.0001 and a sample size of 16. A learning rate decay strategy is used during the fine-tuning training to obtain a well-trained transfer ventilator parameter prediction model for the target patient group. Based on the basic ventilator parameter assessment model, a transfer ventilator parameter assessment model for each patient group is trained using standardized respiratory data time-series sequences. Specific steps include: S411. Freeze the first LSTM layer, second LSTM layer, attention layer, fully connected layer, and Dropout layer in the well-trained basic ventilator parameter evaluation model; S412. Train the frozen model using a time-series sequence of standardized respiratory data from the target patient population, updating only the parameters of the output layer during training; S413. Unfreeze the attention layer, fully connected layer, and Dropout layer, and fine-tune the unfrozen model using a time-series sequence of standardized respiratory data from the target patient group. The fine-tuning training uses a learning rate of 0.0001 and a sample size of ≤16. A learning rate decay strategy is used during the fine-tuning training to obtain a well-trained transfer ventilator parameter evaluation model for the target patient group.

[0096] Step S5: Obtain the patient status data of the target patient and preprocess it to obtain a standardized time series sequence of patient status data. Determine the patient group to which the target patient belongs, and then perform at least one of the following operations: The standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the expected patient status data at a specified time step after time point t are input into the basic ventilator parameter prediction model or the transfer ventilator parameter prediction model of the corresponding patient group to obtain the suggested values ​​of ventilator parameters at time point t. The standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the proposed values ​​of ventilator parameters at time point t are input into the basic ventilator parameter assessment model or the transfer ventilator parameter assessment model of the corresponding patient group to obtain the expected predicted values ​​of patient status data at a specified time step after time point t.

[0097] Step S6: After outputting the recommended values ​​of ventilator parameters using the basic ventilator parameter prediction model or the transfer ventilator parameter prediction model, the recommended values ​​of ventilator parameters are checked against the clinical safety threshold. When the recommended values ​​of ventilator parameters exceed the safety threshold range, they are adjusted to the closest safety threshold range.

[0098] The following are embodiments of the autonomous generation and evaluation system for ventilator parameters provided in this application. This autonomous generation and evaluation system for ventilator parameters belongs to the same inventive concept as the autonomous generation and evaluation methods for ventilator parameters in the above embodiments. For details not described in detail in the embodiments of the autonomous generation and evaluation system for ventilator parameters, please refer to the embodiments of the autonomous generation and evaluation methods for ventilator parameters described above.

[0099] like Figure 2As shown, the ventilator parameter autonomous generation and evaluation system includes: The multimodal respiratory data acquisition module is used to acquire time-series sequences of multimodal respiratory data from different patient groups; The respiratory data preprocessing module is used to preprocess the time series of multimodal respiratory data to obtain a standardized respiratory data time series. The basic prediction model training module is used to set a patient population as the basic population and train a basic ventilator parameter prediction model using the time series of standardized respiratory data of the basic population. The transfer prediction model training module is used to train the transfer ventilator parameter prediction model for each patient group based on the basic ventilator parameter prediction model and using the time series of standardized respiratory data for each patient group. The suggested value generation module is used to input the standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the expected patient status data at a specified time step after time point t into the basic ventilator parameter prediction model or the transfer ventilator parameter prediction model of the corresponding patient group to obtain the suggested ventilator parameter values ​​at time point t. The prediction value generation module is used to input the standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the proposed values ​​of ventilator parameters at time point t into the basic ventilator parameter assessment model or the transferred ventilator parameter assessment model of the corresponding patient group to obtain the expected predicted values ​​of patient status data at a specified time step after time point t.

[0100] The ventilator parameter autonomous generation and evaluation system in this embodiment is used to realize the autonomous generation and evaluation method of ventilator parameters.

[0101] This application also provides an electronic device for implementing the various embodiments of this application. Figure 3 To illustrate the hardware structure of an electronic device according to various embodiments of this application, as shown in the following diagram... Figure 3 As shown, the electronic device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor.

[0102] Those skilled in the art will understand that the electronic device structure involved in the embodiments of this application does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0103] In embodiments of this application, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0104] In this application embodiment, the processor can be implemented using at least one of an Application-Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a processor, a controller, a microcontroller, a microprocessor, or an electronic unit designed to perform the functions described herein. In some cases, such implementations can be implemented within a controller. For software implementations, implementations such as processes or functions can be implemented with separate software modules that allow the performance of at least one function or operation. The software code can be implemented by a software application (or program) written in any suitable programming language, and the software code can be stored in memory and executed by the controller.

[0105] In addition, the electronic device includes some functional modules not shown, which will not be described in detail here.

[0106] Those skilled in the art will understand that the various aspects of the electronic device provided in this application can be implemented as a system, method, or program product. Therefore, the various aspects of this application can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0107] This application also provides a storage medium storing a program product capable of implementing a method for autonomous generation and evaluation of ventilator parameters. In some possible implementations, various aspects of this application can also be implemented as a program product comprising program code that, when run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this application.

[0108] The storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example,, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0109] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for autonomous generation and evaluation of ventilator parameters, characterized in that, include: S1. Classify patient groups and obtain time-series of multimodal respiratory data from different patient groups during ventilator use. The multimodal respiratory data includes patient status data and ventilator parameters. S2. Preprocess the time series of multimodal respiratory data to obtain a standardized respiratory data time series, including a standardized patient status data time series and a standardized ventilator parameter time series; S3. Set a patient group as the base group, and train a base ventilator parameter prediction model using the standardized respiratory data time series of the base group. The base ventilator parameter prediction model is built based on an LSTM network with an attention mechanism. The input includes the standardized patient state data time series within a specified time window ending at the baseline time point and the standardized patient state data at a specified time step after the baseline time point. The output is the suggested value of the ventilator parameter at the baseline time point. During the training of the basic ventilator parameter prediction model, the actual values ​​of ventilator parameters at the baseline time point are extracted from the standardized respiratory parameter time series as true labels. The loss is calculated by comparing the suggested values ​​of ventilator parameters output by the basic ventilator parameter prediction model with the true labels, and the model parameters are updated using the backpropagation algorithm, thereby establishing a mapping relationship between patient status data and ventilator parameters. The baseline ventilator parameter assessment model is trained using a time series of standardized respiratory data from the baseline population. The baseline ventilator parameter assessment model is built on an LSTM network with an attention mechanism. The input includes a time series of standardized patient status data within a specified time window ending at the baseline time point and the proposed values ​​of ventilator parameters at the baseline time point. The output is the predicted value of standardized patient status data at a specified time step after the baseline time point. During the training of the basic ventilator parameter assessment model, the actual values ​​of standardized patient status data extracted from the standardized respiratory parameter time series at a specified time step after the baseline time point are used as the true labels. The loss is calculated by comparing the predicted values ​​of standardized patient status data output by the basic ventilator parameter assessment model with the true labels, and the model parameters are updated using the backpropagation algorithm, thereby establishing a mapping relationship between patient status data and ventilator parameters. S4. Based on the basic ventilator parameter prediction model, the transfer ventilator parameter prediction model for each patient group is trained using the time series of standardized respiratory data for each patient group. Based on the basic ventilator parameter assessment model, the transfer ventilator parameter assessment model for each patient group is trained using time-series standardized respiratory data of each patient group. S5. Obtain the patient status data of the target patient and preprocess it to obtain a standardized time series sequence of patient status data. Determine the patient group to which the target patient belongs, and then perform at least one of the following operations: The standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the expected patient status data at a specified time step after time point t are input into the basic ventilator parameter prediction model or the transfer ventilator parameter prediction model of the corresponding patient group to obtain the suggested values ​​of ventilator parameters at time point t. The standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the proposed values ​​of ventilator parameters at time point t are input into the basic ventilator parameter assessment model or the transfer ventilator parameter assessment model of the corresponding patient group to obtain the expected predicted values ​​of patient status data at a specified time step after time point t.

2. The method for autonomous generation and evaluation of ventilator parameters as described in claim 1, characterized in that, In step S1, the patient group is classified based on at least one of the following factors: race, age group, gender, and disease type.

3. The method for autonomous generation and evaluation of ventilator parameters as described in claim 1, characterized in that, In step S1: Patient status data includes demographic data and physiological and pathological data, among which: Demographic data includes height, weight, and age; Physiological and pathological data include human body pH, PaO2, PaCO2, and body temperature; Ventilator parameters include ventilator settings and ventilator monitoring parameters, among which: Ventilator settings include PEEP, Vt, and RR; Ventilator monitoring parameters include PIP, Pplat, and ventilator flow.

4. The method for autonomous generation and evaluation of ventilator parameters as described in claim 1, characterized in that, In step S3, the basic ventilator parameter prediction model includes an input layer, a first LSTM layer, a second LSTM layer, an attention layer, a fully connected layer, a dropout layer, and an output layer connected in sequence, wherein: The input layer is used to receive the time sequence of standardized patient status data within a specified time window ending at the baseline time point and the expected standardized patient status data at a specified time step after the baseline time point; The first LSTM layer contains multiple first LSTM neurons, which extract short-term temporal features from the data received by the input layer and output a hidden state containing sequence information. The second LSTM layer contains multiple second LSTM neurons, which are used to receive the hidden states output by the first LSTM layer, extract long-term temporal dependency features, and output an encoded temporal feature vector. The attention layer receives the temporal feature vector, calculates the weights of the features at each time step, summarizes the weighted features, and outputs a context vector. The fully connected layer receives the context vector, performs feature compression and nonlinear transformation, and outputs fused features. The Dropout layer is used to receive fused features. It suppresses overfitting by randomly disconnecting some neurons during the forward propagation phase and outputs a regularized feature vector. The output layer receives the regularized feature vector and generates suggested values ​​for ventilator parameters at the baseline time point through a linear activation function.

5. The method for autonomous generation and evaluation of ventilator parameters as described in claim 1, characterized in that, In step S3, the specific training steps for the basic ventilator parameter prediction model include: S301. Construct a respiratory dataset, which includes multiple sets of samples. Each set of samples includes a standardized patient status data time series within a specified time window ending at the baseline time point, and standardized patient status data at a specified time step after the baseline time point. The actual values ​​of ventilator parameters at the baseline time point are set as the true labels of the sample. The actual values ​​of ventilator parameters are extracted from the standardized respiratory parameter time series. S302. Divide the respiratory dataset into a training set, a test set, and a validation set; S303. Initialize the parameters of the basic ventilator parameter prediction model; S304. Input the training set into the initialized basic ventilator parameter prediction model and output the suggested values ​​of ventilator parameters; S305. Use a loss function to calculate the loss between the actual values ​​of ventilator parameters and the suggested values ​​of ventilator parameters at the baseline time point for each group of samples, and update the network parameters through the backpropagation algorithm; S306. Use the validation set to adjust the hyperparameters until the basic ventilator parameter prediction model converges. Use the test set to evaluate the performance of the basic ventilator parameter prediction model to obtain the trained basic ventilator parameter prediction model.

6. The method for autonomous generation and evaluation of ventilator parameters as described in claim 4, characterized in that, In step S4, the specific steps for training the transfer ventilator parameter prediction model for each patient group based on the basic ventilator parameter prediction model and using the time-series of standardized respiratory data for each patient group include: S401. Freeze the first LSTM layer, second LSTM layer, attention layer, fully connected layer, and Dropout layer in the trained basic ventilator parameter prediction model; S402. Train the frozen model using a time-series sequence of standardized respiratory data from the target patient population, updating only the parameters of the output layer during training; S403. Unfreeze the attention layer, fully connected layer, and Dropout layer, and fine-tune the unfrozen model using a time series of standardized respiratory data from the target patient group to obtain a well-trained transfer ventilator parameter prediction model for the target patient group.

7. The method for autonomous generation and evaluation of ventilator parameters as described in claim 1, characterized in that, It also includes step S6: after outputting the recommended values ​​of ventilator parameters using the basic ventilator parameter prediction model or the transferred ventilator parameter prediction model, the recommended values ​​of ventilator parameters are checked for clinical safety thresholds. When the recommended values ​​of ventilator parameters exceed the safety threshold range, they are adjusted to the closest safety threshold range.

8. A system for autonomous generation and evaluation of ventilator parameters, characterized in that, A method for autonomous generation and evaluation of ventilator parameters as described in any one of claims 1-7, comprising: The multimodal respiratory data acquisition module is used to acquire time-series sequences of multimodal respiratory data from different patient groups; The respiratory data preprocessing module is used to preprocess the time series of multimodal respiratory data to obtain a standardized respiratory data time series. The basic prediction model training module is used to set a patient population as the basic population and train a basic ventilator parameter prediction model using the time series of standardized respiratory data of the basic population. The transfer prediction model training module is used to train the transfer ventilator parameter prediction model for each patient group based on the basic ventilator parameter prediction model and using the time series of standardized respiratory data for each patient group. The suggested value generation module is used to input the standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the expected patient status data at a specified time step after time point t into the basic ventilator parameter prediction model or the transfer ventilator parameter prediction model of the corresponding patient group to obtain the suggested ventilator parameter values ​​at time point t. The prediction value generation module is used to input the standardized patient status data time sequence of the target patient within a specified time window ending at time point t and the proposed values ​​of ventilator parameters at time point t into the basic ventilator parameter assessment model or the transferred ventilator parameter assessment model of the corresponding patient group to obtain the expected predicted values ​​of patient status data at a specified time step after time point t.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes a computer program, it implements the steps of the autonomous generation and evaluation method for ventilator parameters as described in any one of claims 1-7.

10. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the autonomous generation and evaluation method for ventilator parameters as described in any one of claims 1-7.

Images