Blood gas index data prediction method, computer readable storage medium and intelligent equipment
By using a neural network model to combine the measured blood gas index data and sign index timing data, predicting blood gas index data, the problem of insufficient blood gas index data was solved, and the effect and accuracy of the sepsis prediction model was improved.
Patent Information
- Application Number
- CN202411868306.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-18
- Publication Date
- 2025-04-29
AI Technical Summary
In the prior art, there is a problem that the amount of data is small in the collection of blood gas index data, which makes it difficult to obtain blood gas index timing data, affecting the training effect and generalization ability of the sepsis prediction model.
By obtaining measured blood gas index data and related sign index timing data, a pre-trained neural network model is used to predict blood gas index data, and combining the sign index timing data to supplement the missing blood gas index data.
Effectively predicting blood gas index data at different moments, solving the problem of insufficient blood gas index data, improving the training effect and generalization ability of the sepsis prediction model, and improving the accuracy of sepsis-related prediction tasks.
Smart Images
Figure CN119905253A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent medical technology, and specifically provides a blood gas index data prediction method, a computer-readable storage medium, and an intelligent device. Background Art
[0002] In the prior art, in order to timely and effectively confirm the patient's sepsis risk, onset time or probability, etc., the patient's relevant indicator data is usually collected, and the sepsis prediction model is trained based on the relevant indicator data to obtain a trained sepsis prediction model, which is used for various sepsis-related prediction tasks. In the related art, when training the sepsis prediction model, the relevant indicator data is mainly based on the blood gas indicator data collected by the arterial blood gas (ABG) analysis method.
[0003] The development of sepsis is a dynamic process, which requires time series data of blood gas indicators. The ABG analysis method is to collect arterial blood-related indicator data of patients with trauma. Two adjacent collections need to be separated by a certain time interval. Correspondingly, there is a small amount of blood gas indicator data, which is not convenient for obtaining blood gas indicator time series data. In addition, the small amount of collected data is not conducive to timely and effective feedback of patient status, affecting the training effect and generalization ability of the sepsis prediction model, and thus affecting the accuracy of sepsis-related prediction tasks. Summary of the invention
[0004] The present application aims to solve the above technical problem, that is, to solve the problem that the existing blood gas index data has a small amount of data and is not convenient for obtaining blood gas index time series data.
[0005] In a first aspect, the present application provides a method for predicting blood gas index data, which comprises:
[0006] Acquire the measured blood gas index data at the first moment and the first vital sign index time series data from the first moment to the first target moment;
[0007] The measured blood gas index data is used as a hidden state and the first vital sign index time series data is used as input, and a pre-trained neural network model is used to predict the first predicted blood gas index data at the first target moment.
[0008] In some embodiments, the method further comprises:
[0009] Acquire time series data of a second vital sign indicator from the first moment to a second target moment;
[0010] The measured blood gas index data is used as a hidden state and the second vital sign index time series data is used as input, and the pre-trained neural network model is used to predict the second predicted blood gas index data at the second target moment.
[0011] In some embodiments, the second target time is greater than the first target time, and the step of acquiring the second vital sign indicator time series data from the first time to the second target time includes:
[0012] The time series data of the vital sign indicators from the first target moment to the second target moment are acquired, and the time series data of the vital sign indicators from the first target moment to the second target moment are merged with the first time series data of the vital sign indicators to obtain the second time series data of the vital sign indicators.
[0013] In some embodiments, the second target time is less than the first target time, and acquiring the second vital sign indicator time series data from the first time to the second target time includes:
[0014] The vital sign indicator time series data from the first moment to the second target moment is filtered out from the first vital sign indicator time series data as the second vital sign indicator time series data.
[0015] In some embodiments, the method further comprises:
[0016] Determine the blood gas index data with the shortest time interval with a third target time from the acquired measured blood gas index data and the predicted blood gas index data as the historical blood gas index data; the third target time is different from the first time and the first target time;
[0017] Acquire the time series data of the third vital sign index from the time point corresponding to the historical blood gas index data to the third target time;
[0018] The historical blood gas index data is used as a hidden state and the third vital sign index time series data is used as input, and the pre-trained neural network model is used to predict the third predicted blood gas index data at the third target moment.
[0019] In some embodiments, the method further comprises obtaining the pre-trained neural network model by the following steps:
[0020] Acquire multiple training data sets, each of which includes measured blood gas index data at a first training moment and time series data of physical sign indicators from the first training moment to a first training target moment, and each of which is pre-marked with measured target blood gas index data at the first training target moment;
[0021] Taking the multiple training data sets as input, using a neural network model to obtain predicted blood gas index data at a first training target moment corresponding to each of the multiple training data sets;
[0022] Based on the predicted blood gas index data and the measured target blood gas index data, the neural network model is iteratively trained until the model converges to obtain the pre-trained neural network model.
[0023] In some embodiments, iteratively training the neural network model includes: in each training cycle, randomly discarding a preset proportion of hidden layer neurons in the neural network model to train the neural network model based on the discarded hidden layer neurons.
[0024] A second aspect of the present application provides a method for predicting the probability of death from sepsis, comprising:
[0025] Acquiring blood gas index time series data, wherein the blood gas index time series data includes blood gas index data at multiple time points, and the blood gas index data at at least some of the multiple time points are obtained by using any of the blood gas index data prediction methods described above;
[0026] The blood gas index time series data is used as input, and a pre-trained sepsis mortality probability prediction model is used to perform prediction to obtain a prediction result.
[0027] In a third aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any of the methods described above is implemented.
[0028] In a fourth aspect, the present application provides a smart device, comprising:
[0029] at least one processor;
[0030] and, a memory communicatively coupled to the at least one processor;
[0031] The memory stores a computer program, and when the computer program is executed by the at least one processor, any of the above-mentioned methods is implemented.
[0032] In the case of adopting the above technical solution, the present application can obtain the measured blood gas index data at the first moment and the first vital sign index time series data from the first moment to the first target moment; take the measured blood gas index data as the hidden state and the first vital sign index time series data as input, and use the pre-trained neural network model to predict the first predicted blood gas index data at the first target moment. This method can effectively predict the blood gas index data at different moments by combining the vital sign index time series data associated with the blood gas index data and using the pre-trained neural network model, solving the problem in the prior art that the amount of measured blood gas index data is small and it is not convenient to obtain the blood gas index time series data. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, in which:
[0034] Figure 1 It is a flowchart of a neural network model training method provided in an embodiment of the present application;
[0035] Figure 2 It is a flow chart of a blood gas index data prediction method provided in an embodiment of the present application;
[0036] Figure 3 It is a flow chart of obtaining the first predicted blood index data based on the gated recurrent neural network provided in an embodiment of the present application;
[0037] Figure 4 is a flow chart of a method for obtaining second predicted blood gas index data provided in an embodiment of the present application;
[0038] Figure 5 is a schematic flow chart of a method for obtaining third predicted blood gas index data provided in an embodiment of the present application;
[0039] Figure 6 It is a flow chart of obtaining the third predicted blood gas index data based on the gated recurrent neural network provided in the embodiment of the present application;
[0040] Figure 7 It is a flow chart of a method for predicting the probability of death from sepsis provided in an embodiment of the present application;
[0041] Figure 8 It is a schematic diagram of the structure of the smart device provided in the embodiment of the present application. DETAILED DESCRIPTION
[0042] Some embodiments of the present application are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application and are not intended to limit the protection scope of the present application.
[0043] Based on the description in the background technology section, it can be seen that the ABG analysis method involves traumatically collecting arterial blood-related index data of patients. Two adjacent collections need to be separated by a certain time interval. Correspondingly, the amount of blood gas index data is small, making it inconvenient to obtain blood gas index time series data.
[0044] Based on this, the present application combines the time series data of the physical sign indicators that are easy to collect and associated with the blood gas index, so as to predict the missing blood gas index data through the physical sign indicator time series data. The physical sign indicators associated with the blood gas index may include physical sign indicators that change with the changes in the blood gas index, and / or, whose changes will cause the changes in the blood gas index. For example, when the patient's physical sign indicator blood oxygen decreases, it is often directly manifested as a decrease in the oxygen partial pressure (PaO2) in the blood gas index, and an increase in the carbon dioxide partial pressure (PaCO2), and the increase in PaCO2 will lead to lactic acid accumulation and a decrease in pH. Therefore, by selecting the patient's physical sign indicator time series data combined with the existing measured blood gas index data, the pre-trained neural network model can effectively predict the blood gas index data at different times, which is also conducive to obtaining the blood gas index time series data, improving the training effect and generalization ability of the sepsis prediction model, and the accuracy of sepsis-related prediction tasks.
[0045] In the embodiment of the present application, the neural network model can be pre-trained. Figure 1 As shown, Figure 1 : is a flowchart of a neural network model training method provided in an embodiment of the present application, which may include:
[0046] Step S11: acquiring a plurality of training data sets, each of which includes the measured blood gas index data at the first training moment and the time series data of the physical sign index from the first training moment to the first training target moment, and each of the plurality of training data sets is pre-marked with the measured target blood gas index data at the first training target moment;
[0047] Step S12: taking multiple training data sets as input, and using a neural network model to obtain predicted blood gas index data at the first training target moment corresponding to each of the multiple training data sets;
[0048] Step S13: Based on the predicted blood gas index data and the measured target blood gas index data, the neural network model is iteratively trained until the model converges to obtain a pre-trained neural network model.
[0049] The ABG analysis method can measure key chemical components such as oxygen, carbon dioxide and pH in the patient's arterial blood. In the embodiment of the present application, the measured blood gas index data may include at least one of hemoglobin, alkalosis, glucose, pH, oxygen partial pressure, carbon dioxide partial pressure, lactic acid and hemoglobin.
[0050] The vital sign indicators may be physiological data of the patient that can be collected in real time by medical equipment, and in some embodiments, include at least one of tidal volume, positive end-expiratory pressure, respiratory rate, blood pressure, blood oxygen saturation, body temperature, heart rate, and oxygen concentration.
[0051] In some embodiments, physical sign indicator time series data can be obtained by using physical sign indicator data at multiple consecutive time points between the first training moment and the first training target moment, so as to improve the effectiveness of the prediction.
[0052] In other embodiments, the physical sign index time series data can be obtained by using the physical sign index data of multiple discrete time points between the first training moment and the first training target moment, and the interval between two adjacent time points is preset, so as to reduce the amount of data collection and improve the prediction efficiency while ensuring the prediction effect. For example, the interval between two adjacent time points can be 15 minutes.
[0053] Among them, the first training target time can be greater than the first training time, and the first training target time can be flexibly set according to actual needs.
[0054] In some embodiments, the neural network model may use a recurrent neural network or a gated recurrent neural network. When a gated recurrent neural network is used, its structure is relatively simple, and overfitting can be effectively avoided during training. In addition, it has fewer parameters and higher computational efficiency. It can achieve better results with relatively less training data, which can effectively reduce data processing costs.
[0055] In some embodiments, the input layer of the gated recurrent neural network may be composed of 16 neurons, of which 8 neurons are used to receive the time series data of the vital sign index, and 8 neurons are used to receive the measured blood gas index data at the first training moment, each neuron may correspond to a different type of blood gas index, and the corresponding output layer may include 8 neurons, representing the predicted blood gas index data corresponding to the 8 measured blood gas index data input. The gated recurrent neural network may include a hidden layer consisting of 140 fully connected neurons. In some embodiments, a tanh activation function may be set after the hidden layer, and the model weights may be initialized using a Xavier normal weight initializer.
[0056] In some embodiments, during the training process, the parameters of the neural network model can also be updated based on the loss value of the predicted blood gas index data and the measured target blood gas index data and the rules of the Adam optimizer.
[0057] In some embodiments, iterative training of the neural network model may include: in each training cycle, randomly discarding a preset proportion of hidden layer neurons in the neural network model, so as to train the neural network model based on the discarded hidden layer neurons. This method can effectively prevent model overfitting.
[0058] See also Figure 2 As shown, Figure 2: is a flow chart of a blood gas index data prediction method provided in an embodiment of the present application, which can be implemented based on the above-mentioned pre-trained neural network model, and the method includes:
[0059] Step S21: acquiring the measured blood gas index data at the first moment and the first vital sign index time series data from the first moment to the first target moment;
[0060] Step S22: using the measured blood gas index data as a hidden state and the first vital sign index time series data as input, using a pre-trained neural network model to predict the first predicted blood gas index data at the first target moment.
[0061] In an embodiment of the present application, the measured blood gas index data may include at least one of hemoglobin, alkalosis, glucose, pH, oxygen partial pressure, carbon dioxide partial pressure, lactic acid and hemoglobin.
[0062] The first vital sign indicator time series data may include time series data of at least one of tidal volume, positive end-expiratory pressure, respiratory rate, blood pressure, blood oxygen saturation, body temperature, heart rate, and oxygen concentration. In some embodiments, the first vital sign indicator time series data may include vital sign indicator data at multiple time points between the first moment and the first target moment, and the multiple time points may be continuous time points or time points with a preset time interval between two adjacent time points.
[0063] Among them, the first target moment is greater than the first moment, and the first target moment can be flexibly set according to needs.
[0064] In some embodiments, obtaining the measured blood gas index data at the first moment in step S21 may be obtaining the measured blood gas index data at a moment that is relatively short from the first target moment, so as to improve the effectiveness of the prediction.
[0065] Take the implementation of the neural network model using a gated recurrent neural network as an example, see Figure 3 As shown, Figure 3 This is a first prediction blood index data acquisition flow chart provided in the embodiment of the present application, in which the blood gas index data ABG can be measured t0 is the hidden state and the first physical indicator time series data X t0～t1 As input, the gated recurrent neural network is used to predict the first predicted blood gas index data ABG at the first target time t1. t1 The method can effectively solve the problem of small amount of measured blood gas index data in the prior art, and is conducive to obtaining blood gas index time series data.
[0066] In some embodiments, to obtain blood gas index time series data, see Figure 4 As shown, the method may also include:
[0067] Step S41: acquiring time series data of a second vital sign indicator from a first moment to a second target moment;
[0068] Step S42: using the measured blood gas index data as a hidden state and the second vital sign index time series data as input, using a pre-trained neural network model to predict the second predicted blood gas index data at the second target moment.
[0069] It should be noted that steps S41 and S42 may be performed simultaneously with steps S21 and S22, or may be performed after step S22.
[0070] The second target time is different from the first target time, and the physical sign indicator type corresponding to the second physical sign indicator time series data may be the same as the physical sign indicator type corresponding to the first physical sign indicator time series data.
[0071] In some embodiments, the second target time is greater than the first target time, and step S41 may be specifically as follows:
[0072] The vital sign indicator time series data from the first target moment to the second target moment is acquired, and the vital sign indicator time series data from the first target moment to the second target moment is merged with the first vital sign indicator time series data to obtain the second vital sign indicator time series data.
[0073] In some embodiments, merging the vital sign indicator time series data from the first target moment to the second target moment with the first vital sign indicator time series data can be as follows: arranging the first vital sign indicator time series data and the vital sign indicator time series data from the first target moment to the second target moment in chronological order to obtain the second vital sign indicator time series data from the first moment to the second target moment.
[0074] In other embodiments, the second target time is less than the first target time, and step S41 may be specifically as follows: filtering out the vital sign indicator time series data from the first time to the second target time from the first vital sign indicator time series data as the second vital sign indicator time series data.
[0075] There may be multiple second target moments, and setting multiple second target moments is helpful in obtaining time series data of blood gas indicators.
[0076] In other embodiments, see Figure 5 As shown, the method may also include:
[0077] Step S51: determining the blood gas index data with the shortest time interval with the third target time from the acquired measured blood gas index data and the predicted blood gas index data as the historical blood gas index data; the third target time is different from the first time and the first target time;
[0078] Step S52: obtaining the third vital sign index time series data from the time point corresponding to the historical blood gas index data to the third target time;
[0079] Step S53: using the historical blood gas index data as a hidden state and the third vital sign index time series data as input, using a pre-trained neural network model to predict the third predicted blood gas index data at the third target moment.
[0080] It should be noted that step S51 may be performed after step S22 or after step S42.
[0081] The acquired measured blood gas index data may include the measured blood gas index data at the first moment, and the acquired predicted blood gas index data may include at least one of the first predicted blood gas index data and the second predicted blood gas index data.
[0082] In some embodiments, step S51 may be to determine the moment with the shortest interval with the third target moment from the first moment, the first target moment and the second target moment, and use the blood gas index data corresponding to the determined moment as the historical blood gas index data.
[0083] When the determined moment is the first moment, step S52 is to obtain the third vital sign index time series data from the first moment to the third target moment, and correspondingly use the measured blood gas index data at the first moment as the historical blood gas index data.
[0084] When the determined moment is the first target moment, step S52 may be to obtain the third vital sign index time series data from the first target moment to the third target moment, and the first predicted blood gas index data is correspondingly used as the hidden state. In some embodiments, when the neural network model adopts a gated recurrent neural network, see Figure 6 As shown, Figure 6 The third predicted blood gas index data acquisition flow chart based on the gated recurrent neural network provided in the embodiment of the present application, wherein the first predicted blood gas index data ABG t1 Afterwards, the first predicted blood gas index data ABG t1 is the hidden state and the third physical sign index time series data is input X t1～t3 , using the pre-trained gated recurrent neural network to predict the third predicted blood gas index data ABG at the third target moment t3 , based on ABG t0 ,ABG t1 and ABG t3 Time series data of blood gas indicators can be obtained.
[0085] When the determined moment is the second target moment, step S52 may be to obtain the third vital sign indicator time series data from the second target moment to the third target moment, and correspondingly the second predicted blood gas indicator data is in a hidden state.
[0086] Among them, the sign indicator type of the third sign indicator time series data can be the same as or different from the sign indicator type of the first sign indicator time series data, and the number of sign indicator types of the third sign indicator time series data can be the same as or different from the number of sign indicator types of the first sign indicator time series data.
[0087] The third target time may be the same as or different from the second target time.
[0088] In the embodiment of the present application, for any target moment, when predicting the blood gas index data of the corresponding target moment based on the blood gas index data prediction method provided in the embodiment of the present application, the blood gas index data of the current target moment can be predicted directly based on the measured blood gas index data with the shortest time interval as the hidden state, or the blood gas index data at the time point with the shortest time interval with the current target moment can be determined in advance from the existing measured blood gas index data, or the existing measured blood gas index data and the existing predicted blood gas index data, and the blood gas index data is input into the neural network model as the hidden state. Accordingly, the time series data of the physical sign index between the time point corresponding to the blood gas index data as the hidden state and the current target moment is used as input, and the predicted blood gas index data of the current target moment is predicted using the pre-trained neural network model to effectively improve the accuracy of the prediction.
[0089] Another aspect of the present application is to provide a method for predicting the probability of death from sepsis, see Figure 7 As shown, Figure 7 : is a flow chart of a method for predicting the probability of death from sepsis provided in an embodiment of the present application, which may include:
[0090] Step S71: obtaining blood gas index time series data, the blood gas index time series data including blood gas index data at multiple time points, the blood gas index data at least part of the multiple time points being obtained by using the blood gas index data prediction method provided by any of the above embodiments of the present application;
[0091] Step S72: using the blood gas index time series data as input, using the pre-trained sepsis mortality probability prediction model to perform prediction and obtain a prediction result.
[0092] In some embodiments, step S71 may specifically include: obtaining the patient's measured blood gas index data at at least one time point and the predicted blood gas index data at multiple time points obtained based on the blood gas index data prediction method provided in the embodiments of the present application.
[0093] In some embodiments, the sepsis mortality probability prediction model can be implemented using a gated recurrent neural network or other models in the art that can achieve the same function. The gated recurrent neural network has a simple structure and requires less data, which can effectively reduce the model construction cycle and reduce the amount of data processing. While ensuring the prediction effect, it can also reduce the cost of early data cleaning.
[0094] The sepsis death probability prediction model can use the same network architecture as the gated recurrent neural network applied to the blood gas index data prediction method in the above-mentioned embodiment of the present application after adaptively adjusting the input layer and the output layer. In some embodiments, the input layer of the sepsis death probability prediction model can be composed of 8 neurons, each neuron is used to receive blood gas index time series data of different types of blood gas indicators of patients.
[0095] In the embodiment of the present application, the trained sepsis mortality probability prediction model can also be obtained by the following steps:
[0096] Acquire a blood gas index training time series data set, the blood gas index training time series data set includes a plurality of blood gas index training time series data, each blood gas index training time series data includes blood gas index data at a plurality of time points, and the blood gas index data at at least some of the plurality of time points are obtained by using the blood gas index data prediction method provided by any of the above embodiments of the present application;
[0097] The blood gas index training time series data set was used as input to train the sepsis mortality probability prediction model, and a trained sepsis mortality probability prediction model was obtained.
[0098] In some embodiments, a stratified three-fold cross-validation method may be used in the above training process and repeated multiple times to evaluate the generalization ability of the model and reduce possible biases.
[0099] In some embodiments, multiple blood gas index training time series data can be acquired in advance, and the acquired multiple blood gas index training time series data can be randomly divided into a blood gas index training time series data set and a model verification data set for model training. In some embodiments, the blood gas index training time series data set can account for 66.67% of the multiple blood gas index training time series data, and the model verification data set can account for 33.33% of the multiple blood gas index training time series data.
[0100] In some embodiments, 75% of the blood gas index training time series data in the blood gas index training time series data set can be used to optimize the weight of the sepsis mortality probability prediction model, and the other 25% of the data is used to adjust the hyperparameters of the sepsis mortality probability prediction model, such as learning rate, network depth and hidden layer size.
[0101] In some embodiments, the trained sepsis mortality probability prediction model can also be evaluated based on the validation data set by using at least one of the following methods: positive predictive value (PPV), negative predictive value (NPV), sensitivity and specificity, area under the ROC curve AUC (Area Under the Receiver Operating Characteristic Curve), area under the precision-recall curve (Area Under the Precision-Recall Curve, AUPRC) and Matthews Correlation Coefficient (MCC).
[0102] By constructing blood gas index training time series data with measured blood gas index data and predicted blood gas index data, the problem of insufficient blood gas index data can be effectively avoided, the training effect and generalization ability of the sepsis prediction model can be improved, and the accuracy of sepsis-related prediction tasks can be improved.
[0103] It is understood by those skilled in the art that all or part of the processes in the method for implementing the above-mentioned embodiments of the present application can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium, etc. that can carry the computer program code.
[0104] Another aspect of the present application further provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the method in any of the above embodiments can be implemented. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present application is a non-temporary computer-readable storage medium.
[0105] In another aspect of the present application, a smart device is provided.
[0106] In an embodiment of an intelligent device according to the present application, the intelligent device may include at least one processor; and a memory connected to the at least one processor in communication; wherein a computer program is stored in the memory, and when the computer program is executed by the at least one processor, the method described in any of the above embodiments is implemented. The intelligent device described in the present application may include a robot and other devices. Figure 8 , Figure 8 8 shows by way of example that the memory 81 and the processor 82 are communicatively connected via a bus.
[0107] In some embodiments of the present application, the smart device may further include at least one sensor, which is used to sense information, such as obtaining body temperature or heart rate. The sensor is communicatively connected to any type of processor mentioned in the present application. The processor communicates with the sensor and / or the autonomous driving system to complete the method described in any of the above embodiments.
[0108] The relevant patient personal information that may be involved in the embodiments of this application is strictly in accordance with the requirements of laws and regulations, following the principles of legality, legitimacy and necessity, based on the reasonable purposes of business scenarios, to process the personal information that patients actively provide during the use of products / services or generated due to the use of products / services, as well as the personal information obtained with the patients' authorization.
[0109] This application attaches great importance to the security of patient personal information and has taken reasonable and feasible security protection measures that meet industry standards to protect patient information and prevent personal information from being accessed, disclosed, used, modified, damaged or lost without authorization.
[0110] So far, the technical solutions of the present application have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments. Without departing from the principles of the present application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present application.
Claims
1. A method for predicting blood gas index data, characterized in that: include: Acquire the measured blood gas index data at the first moment and the first vital sign index time series data from the first moment to the first target moment; The measured blood gas index data is used as a hidden state and the first vital sign index time series data is used as input, and a pre-trained neural network model is used to predict the first predicted blood gas index data at the first target moment.
2. The method according to claim 1, characterized in that: The method further comprises: Acquire time series data of a second vital sign indicator from the first moment to a second target moment; The measured blood gas index data is used as a hidden state and the second vital sign index time series data is used as input, and the pre-trained neural network model is used to predict the second predicted blood gas index data at the second target moment.
3. The method according to claim 2, characterized in that The second target time is greater than the first target time, and the acquiring the second vital sign indicator time series data from the first time to the second target time includes: The time series data of the vital sign indicators from the first target moment to the second target moment are acquired, and the time series data of the vital sign indicators from the first target moment to the second target moment are merged with the first time series data of the vital sign indicators to obtain the second time series data of the vital sign indicators.
4. The method according to claim 2, characterized in that: The second target time is less than the first target time, and acquiring the second vital sign indicator time series data from the first time to the second target time includes: The vital sign indicator time series data from the first moment to the second target moment is filtered out from the first vital sign indicator time series data as the second vital sign indicator time series data.
5. The method according to any one of claims 1 to 4, characterized in that The method further comprises: Determine the blood gas index data with the shortest time interval with a third target time from the acquired measured blood gas index data and the predicted blood gas index data as the historical blood gas index data; the third target time is different from the first time and the first target time; Acquire the time series data of the third vital sign index from the time point corresponding to the historical blood gas index data to the third target time; The historical blood gas index data is used as a hidden state and the third vital sign index time series data is used as input, and the pre-trained neural network model is used to predict the third predicted blood gas index data at the third target moment.
6. The method according to claim 1, characterized in that The method further comprises obtaining the pre-trained neural network model by the following steps: Acquire multiple training data sets, each of which includes measured blood gas index data at a first training moment and time series data of physical sign indicators from the first training moment to a first training target moment, and each of which is pre-marked with measured target blood gas index data at the first training target moment; Taking the multiple training data sets as input, using a neural network model to obtain predicted blood gas index data at a first training target moment corresponding to each of the multiple training data sets; Based on the predicted blood gas index data and the measured target blood gas index data, the neural network model is iteratively trained until the model converges to obtain the pre-trained neural network model.
7. The method according to claim 6, characterized in that Iterative training of the neural network model includes: in each training cycle, randomly discarding a preset proportion of hidden layer neurons in the neural network model to train the neural network model based on the discarded hidden layer neurons.
8. A method for predicting the probability of death from sepsis, characterized in that: include: Acquire blood gas index time series data, wherein the blood gas index time series data includes blood gas index data at multiple time points, and the blood gas index data at at least some of the multiple time points are obtained by using the blood gas index data prediction method described in any one of claims 1 to 7; The blood gas index time series data is used as input, and a pre-trained sepsis mortality probability prediction model is used to perform prediction to obtain a prediction result.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.
10. A smart device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores a computer program, and when the computer program is executed by the at least one processor, the method according to any one of claims 1 to 8 is implemented.