Septicaemia prediction model training method and device, and septicemia prediction method and device

Abstract

The invention provides a septicemia prediction model training method, a septicemia prediction method and a septicemia prediction device, which can be applied to the fields of medical informatics and artificial intelligence. The method comprises the steps that sample time sequence index data is obtained, the sample time sequence index data comprises sample target time sequence index data related to a sepsis patient and sample reference time sequence index data related to a non-sepsis patient, and the sample time sequence index data comprises a plurality of sample time sequence indexes; dividing the sample time sequence index data into multiple segments of sample accumulation time sequence index data, wherein at least two segments of sample accumulation time sequence index data have an overlapping time range; for each segment of sample accumulation time sequence index data, obtaining sample statistical characteristics of a sample time sequence index; determining N sample target time sequence indexes from the plurality of sample time sequence indexes based on kernel principal component analysis, wherein N is a positive integer greater than 0; and training the initial prediction model by using the sample statistical characteristics and the N sample target time sequence indexes to obtain a target prediction model.

Description

Technical Field

[0001] The present disclosure relates to the fields of medical informatics and artificial intelligence, and in particular to a sepsis prediction model training method, a prediction method, and a device. Background Art

[0002] Sepsis is a severe systemic infection caused by pathogens such as bacteria, leading to uncontrolled systemic inflammation and multiple organ failure. Currently, sepsis has become a major challenge facing global public health. Its incidence is extremely high, reaching 535 cases per 100,000 people and is continuing to rise. Therefore, timely and accurate prediction of sepsis development not only facilitates early intervention and prevents progression of the disease, but also effectively reduces waste of medical resources and improves treatment outcomes. Summary of the Invention

[0003] In view of the above problems, the present disclosure provides a sepsis prediction model training method, a prediction method and an apparatus.

[0004] According to a first aspect of the present disclosure, a sepsis prediction model training method, a prediction method, and an apparatus are provided, comprising: obtaining sample time series indicator data, the sample time series indicator data including sample target time series indicator data related to sepsis patients and sample baseline time series indicator data related to non-sepsis patients, the sample time series indicator data including multiple sample time series indicators; dividing the sample time series indicator data into multiple segments of sample cumulative time series indicator data, at least two segments of sample cumulative time series indicator data having overlapping time ranges; obtaining sample statistical features of the sample time series indicators for each segment of sample cumulative time series indicator data; determining N sample target time series indicators from the multiple sample time series indicators based on kernel principal component analysis, where N is a positive integer greater than 0; and training an initial prediction model using the sample statistical features and the N sample target time series indicators to obtain a target prediction model.

[0005] According to an embodiment of the present disclosure, the sample timing indicator data is divided into multiple segments of sample cumulative timing indicator data, including: determining the time length and the number of segments of the sample timing indicator data; processing the time length and the number of segments based on the division rules to obtain multiple segments of sample cumulative timing indicator data, and the initial moments of the multiple segments of sample cumulative timing indicator data are the same.

[0006] According to an embodiment of the present disclosure, the sample time series indicators include static indicators and dynamic indicators. The static indicators include age, and the dynamic indicators include systolic blood pressure, diastolic blood pressure, pH value, blood oxygen saturation, body temperature, heart rate, arterial carbon dioxide partial pressure, respiratory rate and white blood cell count.

[0007] According to an embodiment of the present disclosure, the initial prediction model includes a base learner and a meta-learner, wherein the initial prediction model is trained using sample statistical features and sample target time series indicators to obtain a target prediction model, including: determining the number and type of base learners; for each segment of sample cumulative time series indicator data, the sample statistical features corresponding to the dynamic indicators, the static features corresponding to the static indicators, and the sample target time series indicators are input into each base learner, and the sample prediction probability is output; a sample prediction matrix is determined based on multiple sample prediction probabilities, wherein each row of the prediction matrix represents the prediction probability obtained by all base learners trained based on the same sample cumulative time series data, and each column of the prediction matrix represents the prediction probability obtained by the same base learner trained based on different sample cumulative time series data; the sample prediction matrix is processed by the meta-learner to obtain a sample prediction result; the initial prediction model is trained using the sample prediction result and the label data corresponding to the sample prediction result to obtain a target prediction model.

[0008] According to an embodiment of the present disclosure, a meta-learner is used to process a sample prediction matrix to obtain a sample prediction result, including: horizontally splicing the sample prediction matrix to obtain a splicing feature, where the horizontal splicing representation splices multiple rows of sample prediction probabilities in the sample prediction matrix into at least one row; and using a meta-learner to process the splicing feature to obtain a sample prediction result.

[0009] According to an embodiment of the present disclosure, an initial prediction model is trained using sample prediction results and label data corresponding to the sample prediction results to obtain a target prediction model, including: processing the sample prediction results and label data according to a loss function to obtain a loss function value; and training the initial prediction model according to the loss function value to obtain a target prediction model.

[0010] A second aspect of the present disclosure provides a sepsis prediction method, characterized in that it includes: obtaining time series indicator data of a tester, the time series indicator data including multiple time series indicators; dividing the time series indicator data into multiple segments of cumulative time series indicator data, at least two segments of cumulative time series indicator data having overlapping time ranges; obtaining statistical features of the time series indicator for each segment of cumulative time series indicator data; determining N target time series indicators from the multiple time series indicators based on kernel principal component analysis, where N is a positive integer greater than 0; and processing the statistical features and the N target time series indicators using a target prediction model to obtain a prediction result, wherein the target prediction model is obtained according to the training method according to any one of claims 1 to 6.

[0011] A third aspect of the present disclosure provides a sepsis prediction model training device, comprising: a sample acquisition module for acquiring sample time series indicator data, the sample time series indicator data including sample target time series indicator data related to sepsis patients and sample baseline time series indicator data related to non-sepsis patients, the sample time series indicator data including multiple sample time series indicators; a sample division module for dividing the sample time series indicator data into multiple segments of sample cumulative time series indicator data, at least two segments of sample cumulative time series indicator data having overlapping time ranges; a sample statistical feature acquisition module for acquiring sample statistical features of the sample time series indicators for each segment of sample cumulative time series indicator data; a sample target time series indicator determination module for determining N sample target time series indicators from multiple sample time series indicators based on kernel principal component analysis, where N is a positive integer greater than 0; and a training module for training an initial prediction model using the sample statistical features and the N sample target time series indicators to obtain a target prediction model.

[0012] A fourth aspect of the present disclosure provides a sepsis prediction device, comprising: an acquisition module for acquiring time series indicator data of a tester, the time series indicator data including multiple time series indicators; a division module for dividing the time series indicator data into multiple segments of cumulative time series indicator data, at least two segments of cumulative time series indicator data having overlapping time ranges; a statistical feature acquisition module for acquiring statistical features of the time series indicator for each segment of cumulative time series indicator data; a target time series indicator determination module for determining N target time series indicators from the multiple time series indicators based on kernel principal component analysis, where N is a positive integer greater than 0; and a prediction result acquisition module for processing the statistical features and the N target time series indicators using a target prediction model to obtain a prediction result.

[0013] A fifth aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0014] The sixth aspect of the present disclosure further provides a computer program product, comprising a computer program or instructions, which implement the steps of the above method when executed by a processor.

[0015] According to an embodiment of the present disclosure, by dividing the sample time series indicator data into multiple segments of sample cumulative time series indicator data, at least two segments of the sample cumulative time series indicator data have overlapping time ranges, for each segment of sample cumulative time series indicator data, the sample statistical characteristics of the sample time series indicator are obtained, and based on the kernel principal component analysis, N sample target time series indicators are determined from the multiple sample time series indicators, thereby performing dimensionality reduction and feature extraction on the sample cumulative time series indicator data, and using the sample statistical characteristics and the sample target time series indicator characteristics to train the initial prediction model to obtain the target prediction model. Compared with traditional methods that usually only use data of a single length to directly train the model, the present disclosure divides the sample time series indicator data into multiple sample cumulative time series indicator data of different lengths, so that the model can capture more information, greatly expands the feature space that the model can access, and improves the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0017] Figure 1 Schematically illustrates an application scenario diagram of a sepsis prediction model training method and a prediction method according to an embodiment of the present disclosure;

[0018] Figure 2 The flowchart of the sepsis prediction model training method according to an embodiment of the present disclosure is schematically shown;

[0019] Figure 3 Schematic diagram showing example time series of five parameters in a sepsis patient;

[0020] Figure 4 The training process of the sepsis prediction model is schematically shown;

[0021] Figure 5 The flowchart of the sepsis prediction method according to an embodiment of the present disclosure is schematically shown;

[0022] Figure 6 Schematically shows a structural block diagram of a sepsis prediction model training device according to an embodiment of the present disclosure;

[0023] Figure 7 Schematically shows a structural block diagram of a sepsis prediction model prediction device according to an embodiment of the present disclosure;

[0024] Figure 8 A block diagram of an electronic device suitable for implementing a sepsis prediction model training method and a prediction method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0025] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0026] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0027] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0028] When expressions such as "at least one of A, B, and C, etc." are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0029] Sepsis is a severe systemic infection caused by pathogens such as bacteria, leading to uncontrolled systemic inflammation and multiple organ failure. Currently, sepsis has become a major challenge facing global public health. Its incidence is extremely high, reaching 535 cases per 100,000 people and showing a continuously increasing trend. The in-hospital mortality rate is as high as 25%-30%. Therefore, timely and accurate prediction of sepsis can not only facilitate early intervention measures and prevent disease progression, but also effectively reduce the waste of medical resources and improve treatment outcomes. However, existing sepsis prediction models have shortcomings in processing time series information, model fusion, and information integration, which affect the accuracy of sepsis prediction.

[0030] In view of this, the present disclosure provides a sepsis prediction model training method, prediction method, and device. The method includes: obtaining sample time series indicator data, the sample time series indicator data including sample target time series indicator data related to sepsis patients and sample baseline time series indicator data related to non-sepsis patients, and the sample time series indicator data including multiple sample time series indicators; dividing the sample time series indicator data into multiple segments of sample cumulative time series indicator data, at least two segments of sample cumulative time series indicator data having overlapping time ranges; obtaining sample statistical features of the sample time series indicators for each segment of sample cumulative time series indicator data; determining N sample target time series indicators from the multiple sample time series indicators based on kernel principal component analysis, where N is a positive integer greater than 0; and training an initial prediction model using the sample statistical features and the N sample target time series indicators to obtain a target prediction model.

[0031] In the technical solutions disclosed herein, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0032] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided by the embodiments of the present disclosure all provide users with corresponding operation portals for them to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge, and skills, and have reached a certain level of professionalism.

[0033] Figure 1 The following schematically illustrates an application scenario diagram of a sepsis prediction model training method and a prediction method according to an embodiment of the present disclosure.

[0034] like Figure 1As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or optical fiber cables.

[0035] A user may use a first terminal device 101, a second terminal device 102, or a third terminal device 103 to interact with a server 105 via a network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, or the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (for example only).

[0036] The first terminal device 101 , the second terminal device 102 , and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0037] The server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process received data such as user requests, and feed back processing results (e.g., web pages, information, or data obtained or generated based on user requests) to the terminal devices.

[0038] It should be noted that the sepsis prediction model training method and prediction method provided in the embodiments of the present disclosure can generally be executed by the server 105. Accordingly, the sepsis prediction model training device and prediction device provided in the embodiments of the present disclosure can generally be set in the server 105. The sepsis prediction model training method and prediction method provided in the embodiments of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the sepsis prediction model training device and prediction device provided in the embodiments of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0039] It should be understood that Figure 1The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0040] Figure 2 The flowchart of the sepsis prediction model training method according to an embodiment of the present disclosure is schematically shown.

[0041] like Figure 2 As shown, the sepsis prediction model training method of this embodiment includes operations S210 to S240, and the sepsis prediction model training method can be executed by an electronic device.

[0042] In operation S210 , sample timing indicator data is acquired, where the sample timing indicator data includes a plurality of sample timing indicators.

[0043] In operation S220 , the sample time series indicator data is divided into multiple segments of sample accumulated time series indicator data, and at least two segments of the sample accumulated time series indicator data have overlapping time ranges.

[0044] In operation S230 , time series indicator data is accumulated for each sample segment to obtain sample statistical features of the sample time series indicator.

[0045] In operation S240 , N sample target timing indices are determined from the plurality of sample timing indices based on kernel principal component analysis.

[0046] In operation S250 , an initial prediction model is trained using sample statistical features and N sample target time series indicators to obtain a target prediction model.

[0047] According to an embodiment of the present disclosure, the sample time series indicator data may include sample target time series indicator data related to sepsis patients and sample baseline time series indicator data related to non-sepsis patients. The sample target time series indicator data and the sample baseline time series indicator data may include the same sample time series indicators, such as age, blood oxygen saturation, etc., but are not limited to this. The embodiments of the present disclosure do not limit the categories of sample time series indicators.

[0048] According to an embodiment of the present disclosure, a roughly equal proportion of sepsis patients and non-sepsis patients can be extracted from a hospital patient information database. All patients meet the following conditions: they are 18 years of age or older and have been admitted to an intensive care unit (ICU). Each patient's record can include sample time series indicators related to sepsis. Each sample time series indicator is averaged hourly to generate hourly records. During the actual data collection process, some missing values may exist. To ensure data integrity, multiple imputation can be used to fill missing values to avoid affecting subsequent analysis.

[0049] According to an embodiment of the present disclosure, assuming that the time required to predict the onset of sepsis is r hours in advance (r is a non-negative integer, and this period is called the "prediction time"), and the length of the time series to be used is determined to be l (a positive integer) hours, then time series data of the corresponding length must be extracted from the patient's complete record for subsequent analysis and prediction.

[0050] Figure 3 Schematic diagram showing example time series of five parameters for a sepsis patient.

[0051] like Figure 3 As shown, roughly equal proportions of sepsis patients and non-sepsis patients can be extracted from the MIMIC IV database. All patients meet the following conditions: they are 18 years old or older and have been admitted to the Intensive Care Unit (ICU). Time series indicator data of the patients' temperature, heart rate, respiratory rate, arterial carbon dioxide partial pressure, and white blood cell count were collected. The average value of each indicator was calculated hourly, and the missing values were filled using the multiple imputation method. , extract the corresponding time series of length 10 hours from the complete records of sepsis patients and non-sepsis patients respectively. Figure 3 The figure shows the time series of five parameters for a sepsis patient. Using the gold standard established by those skilled in the art, the onset of sepsis (sepsis onset) can be determined. For sepsis patients, only the 10 hours of time series prior to (onset - 3) hours were extracted. For non-septic patients, a random 10-hour time series was selected from the entire recording period.

[0052] First, for patients with sepsis, it is necessary to determine the time when they develop sepsis. The gold standard proposed by those skilled in the art can be used, which consists of two criteria: (1) the patient suffers from sepsis; (2) the patient exhibits systemic inflammatory response syndrome (SIRS) for at least 5 consecutive hours, referred to as 5h-SIRS. The specific definition of the conditions for suffering from SIRS is shown in Table 1. For patients who meet these two conditions, the first hour of the 5h-SIRS can be used as the time when the patient develops sepsis (called the "onset" moment). The goal of this disclosure is to predict sepsis r hours in advance, so the time period from (onset-r) hours to onset will not be considered by the model. For patients with sepsis, only the time series of 1 hour before (onset-r) hours needs to be extracted.

[0053] For non-septic patients, a continuous 1-hour time series can be randomly selected from the entire recording interval for subsequent analysis.

[0054] Table 1

[0055]

[0056] According to an embodiment of the present disclosure, sample time series indicator data can be divided into multiple segments of sample cumulative time series indicator data, with at least two segments of sample cumulative time series indicator data having overlapping time ranges. For example, if the time series corresponding to the sample time series indicator data is 0 to 10 hours, it can be divided into 0-4 hours, 3-7 hours, and 6-10 hours.

[0057] According to an embodiment of the present disclosure, for each segment of sample accumulated time series indicator data, the sample statistical characteristics of the sample time series indicator can be calculated, and the sample statistical characteristics can include the maximum value, minimum value, and variance of the sample accumulated time series indicator data.

[0058] For example, the accumulated time series indicator data for each sample segment may include time series indicator data related to temperature, heart rate, respiratory rate, arterial carbon dioxide partial pressure, and white blood cell count. For the time series data of the five dimensions of temperature, heart rate, respiratory rate, arterial carbon dioxide partial pressure, and white blood cell count, the sample statistical characteristics such as the maximum value, minimum value, and variance corresponding to each dimension can be calculated.

[0059] According to the embodiments of the present disclosure, since the data used is a multivariate time series with characteristics such as high dimensionality, noise and redundant features, if the model is trained directly on the sample time series indicator data, it may cause the model to be too complex, thereby affecting the prediction effect. Therefore, preprocessing the extracted sample time series indicator data is the key to effective analysis. Kernel Principal Component Analysis (KPCA) can be used to reduce the dimension and extract features of the data. Except for static indicators, the remaining indicators are dynamic data that changes over time. Each dimension of the time series (i.e., dynamic indicators) is regarded as a variable, and then KPCA is used to reduce the dimension of the dynamic indicators separately, and N main dynamic indicators are extracted therefrom. In this way, each segment of the sample cumulative time series indicator data will generate a feature vector.

[0060] According to an embodiment of the present disclosure, an initial prediction model may be trained using sample statistical features and N sample target time series indicators to obtain a target prediction model.

[0061] According to an embodiment of the present disclosure, by dividing the sample time series indicator data into multiple segments of sample cumulative time series indicator data, at least two segments of the sample cumulative time series indicator data have overlapping time ranges, for each segment of sample cumulative time series indicator data, the sample statistical characteristics of the sample time series indicator are obtained, and based on the kernel principal component analysis, N sample target time series indicators are determined from the multiple sample time series indicators, thereby performing dimensionality reduction and feature extraction on the sample cumulative time series indicator data, and using the sample statistical characteristics and the sample target time series indicator characteristics to train the initial prediction model to obtain the target prediction model. Compared with traditional methods that usually only use data of a single length to directly train the model, the present disclosure divides the sample time series indicator data into multiple sample cumulative time series indicator data of different lengths, so that the model can capture more information, greatly expands the feature space that the model can access, and improves the accuracy of the prediction.

[0062] According to an embodiment of the present disclosure, the sample time series indicators include static indicators and dynamic indicators. The static indicators include age, and the dynamic indicators include systolic blood pressure, diastolic blood pressure, hydrogen ion concentration index (pH) value, blood oxygen saturation, body temperature, heart rate, arterial carbon dioxide partial pressure, respiratory rate and white blood cell count.

[0063] According to an embodiment of the present disclosure, an indicator that can be used to determine whether sepsis is present can be used as a sample time series indicator. The sample time series indicator may include static indicators and dynamic indicators. The static indicator may include age, and the dynamic indicator may include systolic blood pressure, diastolic blood pressure, pH value, blood oxygen saturation, body temperature, heart rate, arterial carbon dioxide partial pressure, respiratory rate and white blood cell count, but is not limited to this. The embodiment of the present disclosure does not limit the specific category of the sample time series indicator.

[0064] According to the embodiments of the present disclosure, it should be noted that during data extraction for non-septic patients, only sample time series indicators of age, systolic blood pressure, diastolic blood pressure, pH value, blood oxygen saturation, body temperature, heart rate, respiratory rate, and white blood cell count may be selected. Arterial carbon dioxide partial pressure and respiratory rate may both be used as criteria for determining whether SIRS is present. For non-septic patients, any one of these may be selected.

[0065] According to an embodiment of the present disclosure, the sample timing indicator data is divided into multiple segments of sample cumulative timing indicator data, including: determining the time length and the number of segments of the sample timing indicator data; processing the time length and the number of segments based on the division rules to obtain multiple segments of sample cumulative timing indicator data, and the initial moments of the multiple segments of sample cumulative timing indicator data are the same.

[0066] According to the embodiments of the present disclosure, the time length and the number of segments of the sample timing indicator data can be determined; based on the division rules, the time length and the number of segments can be processed to obtain the cumulative timing indicator data of multiple segments of samples, and the initial moments of the cumulative timing indicator data of multiple segments of samples can be the same.

[0067] For example, suppose Represents the sample time series indicator data of 1 hour (since the time series used are all retrospective data, they are collectively referred to as the 1-hour lookback window, abbreviated as 1h look back). If the sample time series indicator data is divided into (positive integer and ) The accumulated time series indicator data of the sample segment is divided into the following rules: h lookback, h look back, ..., lh look back. For example, if the selected time length l is 10 hours and it is divided into If the 10-hour lookback period is divided into five segments using the aforementioned division rules, the 10-hour lookback period will be divided into five segments of cumulative time series indicator data: 2-hour lookback, 4-hour lookback, 6-hour lookback, 8-hour lookback, and 10-hour lookback. The initial moments of these different lengths of cumulative time series indicator data are the same. These different lengths of cumulative time series indicator data all contain the patient's most recent data points.

[0068] According to the embodiments of the present disclosure, by ensuring that the cumulative time series indicator data for each sample segment contains the latest data points, the model can reflect the latest data dynamics in real time. This feature not only improves the timeliness of the model but also enables the model to adjust prediction strategies in a timely manner, thereby more effectively responding to rapid changes in the clinical environment. By extracting information from multiple dimensions and time scales, the model can obtain richer and more comprehensive feature representations than traditional methods, thereby improving the accuracy and reliability of predictions.

[0069] According to an embodiment of the present disclosure, the initial prediction model includes a base learner and a meta-learner, wherein the initial prediction model is trained using sample statistical features and sample target time series indicators to obtain a target prediction model, including: determining the number and type of base learners; for each segment of sample cumulative time series indicator data, the sample statistical features corresponding to the dynamic indicators, the static features corresponding to the static indicators, and the sample target time series indicators are input into each base learner, and the sample prediction probability is output; a sample prediction matrix is determined based on multiple sample prediction probabilities, wherein each row of the prediction matrix represents the prediction probability obtained by all base learners trained based on the same sample cumulative time series data, and each column of the prediction matrix represents the prediction probability obtained by the same base learner trained based on different sample cumulative time series data; the sample prediction matrix is processed by the meta-learner to obtain a sample prediction result; the initial prediction model is trained using the sample prediction result and the label data corresponding to the sample prediction result to obtain a target prediction model.

[0070] According to an embodiment of the present disclosure, the initial prediction model includes a base learner and a meta learner. The base learner can select three models: support vector machine, random forest and gradient boosting tree, but is not limited to this. The embodiment of the present disclosure does not limit the number and type of base learners.

[0071] According to the embodiments of the present disclosure, through the combination of multiple base learners, the model can automatically adjust in different scenarios, combining the advantages of each base learner to make up for the limitations of a single base learner.

[0072] According to an embodiment of the present disclosure, for each segment of sample accumulated time series indicator data, the sample statistical features corresponding to the dynamic indicators, the static features corresponding to the static indicators, and the sample target time series indicators are input into each base learner, and the sample prediction probability is output. For example, the static features corresponding to the static indicator "age" and the maximum, minimum, and variance of the remaining dynamic indicators, as well as the feature vectors obtained by KPCA, are input as input features into each base learner, and the sample prediction probability is output.

[0073] According to the embodiment of the present disclosure, the input features obtained by accumulating the time series index data of each sample are input into each base learning machine respectively. Each base learner predicts the cumulative time series indicator data for each sample segment and generates the corresponding sample prediction probability. The prediction results for each patient are organized into a prediction matrix. Each row of the prediction matrix represents the prediction probability obtained by all base learners trained on the same sample cumulative time series data, and each column of the prediction matrix represents the prediction probability obtained by the same base learner trained on different sample cumulative time series data.

[0074] According to an embodiment of the present disclosure, a meta-learner can be used to process a sample prediction matrix to obtain a sample prediction result; the sample prediction result and the label data corresponding to the sample prediction result are used to train an initial prediction model to obtain a target prediction model.

[0075] According to the embodiments of the present disclosure, by dividing the time series into multiple samples of varying lengths, the feature space accessible to the meta-learner is significantly expanded. By combining different base learners with sample-based time series metrics of varying lengths, the model effectively leverages the synergistic effects between these multiple base learners. This synergistic effect can effectively mitigate the potential bias inherent in a single model, thereby improving the accuracy and robustness of the overall prediction.

[0076] According to an embodiment of the present disclosure, a meta-learner is used to process a sample prediction matrix to obtain a sample prediction result, including: horizontally splicing the sample prediction matrix to obtain a splicing feature, where the horizontal splicing representation splices multiple rows of sample prediction probabilities in the sample prediction matrix into at least one row; and using a meta-learner to process the splicing feature to obtain a sample prediction result.

[0077] According to an embodiment of the present disclosure, the sample prediction matrix may be horizontally spliced to obtain a splicing feature, where the horizontal splicing representation splices multiple rows of sample prediction probabilities in the sample prediction matrix into at least one row.

[0078] For example, the prediction matrix , as shown in formula (1):

[0079] (1);

[0080] in Indicates that the i-th patient uses h look back data training base learner The predicted probability is obtained.

[0081] According to the embodiments of the present disclosure, All the predicted probabilities in are horizontally spliced to obtain a length of dimensional feature vector , as shown in formula (2)

[0082] (2);

[0083] According to an embodiment of the present disclosure, a meta-learner may be used to process concatenated features to obtain sample prediction results.

[0084] According to an embodiment of the present disclosure, an initial prediction model is trained using sample prediction results and label data corresponding to the sample prediction results to obtain a target prediction model, including: processing the sample prediction results and label data according to a loss function to obtain a loss function value; and training the initial prediction model according to the loss function value to obtain a target prediction model.

[0085] According to an embodiment of the present disclosure, the meta-learner The concatenated features can be used for training to obtain sample prediction results. The sample prediction results and label data are processed according to the loss function to obtain the loss function value. The goal is to enable the meta-learner to more accurately predict the onset of sepsis by minimizing the loss function, as shown in formula (3). The meta-learner can choose a relatively simple model such as logistic regression or decision tree. The training process is to minimize the error between the sample prediction results and the label data (the label data represents whether sepsis occurs) by continuously adjusting the parameters of the meta-learner:

[0086] (3);

[0087] in, represents the splicing features of the i-th patient, are the parameters of the meta-learner, represents the actual label of the i-th patient, that is, whether sepsis occurs.

[0088] Figure 4 The training process of the sepsis prediction model is schematically shown.

[0089] like Figure 4 As shown, one hour of sample time series indicator data 401 can be obtained, and the one hour of sample time series indicator data can be divided into k segments of sample cumulative time series indicator data. Figure 4 In the example, when the sample time series index data is divided into multiple sections of sample cumulative time series index data, the end time of the multiple sections of sample cumulative time series index data is the same. For each section of sample cumulative time series index data, multiple base learners 402 are used to process it to obtain the sample prediction probability. Base learner You can choose M, that is Based on the multiple sample prediction probabilities, a sample prediction matrix 403 can be determined. Sample prediction matrices 403 are concatenated to obtain K*M concatenated features 404. Concatenated features 404 are processed by a meta-learner 405 to obtain sample prediction results. The sample prediction results and the corresponding labeled data are used to train an initial prediction model to obtain a target prediction model. Training is now complete 406.

[0090] Figure 5 The flowchart of the sepsis prediction method according to an embodiment of the present disclosure is schematically shown.

[0091] like Figure 5 As shown, the sepsis prediction method of this embodiment includes operations S510 to S570, and the sepsis prediction method can be executed by an electronic device.

[0092] In operation S510 , timing indicator data of a tester is acquired, where the timing indicator data includes a plurality of timing indicators.

[0093] In operation S520 , the time series indicator data is divided into multiple segments of accumulated time series indicator data, and at least two segments of accumulated time series indicator data have overlapping time ranges.

[0094] In operation S530 , for each segment of accumulated time series indicator data, statistical features of the time series indicator are obtained.

[0095] In operation S540 , N target timing indices are determined from the plurality of timing indices based on kernel principal component analysis.

[0096] In operation S550 , the target prediction model is used to process the statistical features and the N target time series indicators to obtain a prediction result.

[0097] According to an embodiment of the present disclosure, the tester's timing indicator data is obtained, and the timing indicator data may include multiple timing indicators. The timing indicators need to correspond to the sample timing indicators used in the training of the target prediction model. The timing indicator data can be divided into multiple segments of cumulative timing indicator data, and at least two segments of cumulative timing indicator data have overlapping time ranges. For each segment of cumulative timing indicator data, the statistical characteristics of the timing indicators can be obtained. Based on kernel principal component analysis, N target timing indicators are determined from multiple timing indicators. By using the target prediction model to process the statistical characteristics and N target timing indicators, a prediction result can be obtained.

[0098] For example, when the model receives a new patient data When , the new data will first be preprocessed accordingly, and then input into the base learner for prediction, and a splicing feature can be obtained. , as shown in formula (4):

[0099] (4);

[0100] Then, will be passed as input to the meta-learner By making predictions, we can get the final prediction results, as shown in formula (5):

[0101] (5);

[0102] According to the embodiments of the present disclosure, the obtained prediction results cannot directly lead to the diagnosis results of the disease or the health status, but only serve as intermediate result information, and all the above steps are implemented and executed by computers and other devices.

[0103] Figure 6 The structural block diagram of the sepsis prediction model training device according to an embodiment of the present disclosure is schematically shown.

[0104] like Figure 6 As shown, the sepsis prediction model training device of this embodiment includes a sample acquisition module 610, a sample division module 620, a sample statistical feature acquisition module 630, a sample target time series index determination module 640 and a training module 650.

[0105] The sample acquisition module 610 is used to acquire sample time series indicator data, wherein the sample time series indicator data includes sample target time series indicator data related to sepsis patients and sample baseline time series indicator data related to non-sepsis patients. The sample time series indicator data includes multiple sample time series indicators.

[0106] The sample division module 620 is configured to divide the sample time series indicator data into multiple segments of sample accumulated time series indicator data, where at least two segments of sample accumulated time series indicator data have overlapping time ranges.

[0107] The sample statistical feature acquisition module 630 is used to accumulate time series indicator data for each sample segment and obtain the sample statistical features of the sample time series indicator.

[0108] a sample target timing index determination module 640 , configured to determine N sample target timing indexes from a plurality of sample timing indexes based on kernel principal component analysis, where N is a positive integer greater than 0;

[0109] The training module 650 is used to train the initial prediction model using the sample statistical features and N sample target time series indicators to obtain a target prediction model.

[0110] According to an embodiment of the present disclosure, by dividing the sample time series indicator data into multiple segments of sample cumulative time series indicator data, at least two segments of the sample cumulative time series indicator data have overlapping time ranges, for each segment of sample cumulative time series indicator data, the sample statistical characteristics of the sample time series indicator are obtained, and based on the kernel principal component analysis, N sample target time series indicators are determined from the multiple sample time series indicators, thereby performing dimensionality reduction and feature extraction on the sample cumulative time series indicator data, and using the sample statistical characteristics and the sample target time series indicator characteristics to train the initial prediction model to obtain the target prediction model. Compared with traditional methods that usually only use data of a single length to directly train the model, the present disclosure divides the sample time series indicator data into multiple sample cumulative time series indicator data of different lengths, so that the model can capture more information, greatly expands the feature space that the model can access, and improves the accuracy of the prediction.

[0111] According to an embodiment of the present disclosure, the sample division module includes: a division segment number determination unit and a sample cumulative time series indicator data acquisition unit.

[0112] The segment number determination unit is used to determine the time length and segment number of the sample time series indicator data.

[0113] The sample cumulative time series indicator data obtaining unit is used to process the time length and the number of divided segments based on the division rule to obtain multiple segments of sample cumulative time series indicator data, and the initial moments of the multiple segments of sample cumulative time series indicator data are the same.

[0114] According to an embodiment of the present disclosure, the sample time series indicators include static indicators and dynamic indicators. The static indicators include age, and the dynamic indicators include systolic blood pressure, diastolic blood pressure, pH value, blood oxygen saturation, body temperature, heart rate, arterial carbon dioxide partial pressure, respiratory rate and white blood cell count.

[0115] According to an embodiment of the present disclosure, the initial prediction model includes a base learner and a meta learner.

[0116] According to an embodiment of the present disclosure, the training module includes: a base learner determination unit, a sample prediction probability output unit, a sample prediction matrix determination unit, a sample prediction result acquisition unit and a target prediction model acquisition unit.

[0117] The base learner determination unit is used to determine the number and type of base learners.

[0118] The sample prediction probability output unit is used to accumulate time series indicator data for each sample segment, input the sample statistical features corresponding to the dynamic indicators, the static features corresponding to the static indicators, and the sample target time series indicators into each base learner, and output the sample prediction probability.

[0119] A sample prediction matrix determination unit is used to determine a sample prediction matrix based on multiple sample prediction probabilities, where each row of the prediction matrix represents the prediction probability obtained by all base learners trained based on the same sample cumulative time series data, and each column of the prediction matrix represents the prediction probability obtained by the same base learner trained based on different sample cumulative time series data.

[0120] The sample prediction result obtaining unit is used to process the sample prediction matrix using the meta-learner to obtain the sample prediction result.

[0121] The target prediction model obtaining unit is used to train the initial prediction model using the sample prediction results and the label data corresponding to the sample prediction results to obtain the target prediction model.

[0122] According to an embodiment of the present disclosure, the sample prediction result obtaining unit includes: a splicing feature subunit and a sample prediction result subunit.

[0123] The splicing feature subunit is used to horizontally splice the sample prediction matrix to obtain a splicing feature. The horizontal splicing representation splices multiple rows of sample prediction probabilities in the sample prediction matrix into at least one row.

[0124] The sample prediction result subunit is used to use the meta-learner to process the splicing features and obtain the sample prediction result.

[0125] According to an embodiment of the present disclosure, the training module includes: a loss function value obtaining unit and a training unit.

[0126] The loss function value obtaining unit is used to process the sample prediction results and label data according to the loss function to obtain the loss function value.

[0127] The training unit is used to train the initial prediction model according to the loss function value to obtain the target prediction model.

[0128] Figure 7 The figure schematically shows a structural block diagram of a sepsis prediction device according to an embodiment of the present disclosure.

[0129] like Figure 7 As shown, the sepsis prediction device of this embodiment includes an acquisition module 710 , a division module 720 , a statistical feature acquisition module 730 , a target time series index determination module 740 and a prediction result acquisition module 750 .

[0130] The acquisition module 710 is configured to acquire the tester's timing indicator data, where the timing indicator data includes a plurality of timing indicators.

[0131] The division module 720 is configured to divide the time series indicator data into multiple segments of accumulated time series indicator data, where at least two segments of the accumulated time series indicator data have overlapping time ranges.

[0132] The statistical feature acquisition module 730 is configured to acquire the statistical features of the time series indicator for each segment of accumulated time series indicator data.

[0133] The target timing indicator determination module 740 is configured to determine N target timing indicators from the plurality of timing indicators based on kernel principal component analysis, where N is a positive integer greater than 0.

[0134] The prediction result obtaining module 750 is used to process the statistical features and the N target time series indicators using a target prediction model to obtain a prediction result.

[0135] According to an embodiment of the present disclosure, any multiple modules among the sample acquisition module 610, the sample partitioning module 620, the sample statistical feature acquisition module 630, the sample target time series index determination module 640, the training module 650, the acquisition module 710, the partitioning module 720, the statistical feature acquisition module 730, the target time series index determination module 740, and the prediction result acquisition module 750 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the sample acquisition module 610, the sample partitioning module 620, the sample statistical feature acquisition module 630, the sample target timing indicator determination module 640, the training module 650, the acquisition module 710, the partitioning module 720, the statistical feature acquisition module 730, the target timing indicator determination module 740 and the prediction result acquisition module 750 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or can be implemented in any one of the three implementation methods of software, hardware and firmware, or in an appropriate combination of any of them. Alternatively, at least one of the sample acquisition module 610, the sample partitioning module 620, the sample statistical feature acquisition module 630, the sample target timing indicator determination module 640, the training module 650, the acquisition module 710, the partitioning module 720, the statistical feature acquisition module 730, the target timing indicator determination module 740 and the prediction result acquisition module 750 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0136] Figure 8 A block diagram of an electronic device suitable for implementing a sepsis prediction model training method and a prediction method according to an embodiment of the present disclosure is schematically shown.

[0137] like Figure 8 As shown, the electronic device according to an embodiment of the present disclosure includes a processor 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage unit 808 into a random access memory (RAM) 803. The processor 801 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 801 may also include onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0138] Various programs and data required for the operation of the electronic device are stored in RAM 803. The processor 801, ROM 802, and RAM 803 are connected to each other via a bus 804. The processor 801 performs various operations of the method flow according to the embodiment of the present disclosure by executing the programs in ROM 802 and / or RAM 803. It should be noted that the programs may also be stored in one or more memories other than ROM 802 and RAM 803. The processor 801 may also perform various operations of the method flow according to the embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0139] According to an embodiment of the present disclosure, the electronic device may further include an input / output (I / O) interface 805, which is also connected to the bus 804. The electronic device may further include one or more of the following components connected to the I / O interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 808 including a hard disk; and a communication section 809 including a network interface card such as a LAN card or modem. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. Removable media 811, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in the drive 810 as needed, so that computer programs read from the removable media can be installed in the storage section 808 as needed.

[0140] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when executed, implements the method according to the embodiments of the present disclosure.

[0141] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, and may include, for example, but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 802 and / or RAM 803 described above, and / or one or more memories other than ROM 802 and RAM 803.

[0142] Embodiments of the present disclosure also include a computer program product comprising a computer program containing program code for executing the method shown in the flowchart. When the computer program product is executed in a computer system, the program code causes the computer system to implement the sepsis prediction model training method and prediction method provided in the embodiments of the present disclosure.

[0143] The computer program executes the above functions defined in the system / device of the embodiment of the present disclosure when the computer program is executed by the processor 801. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0144] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 809, and / or installed from a removable medium 811. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0145] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 809, and / or installed from a removable medium 811. When the computer program is executed by the processor 801, the above-described functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0146] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages include, but are not limited to, languages such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0147] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0148] Those skilled in the art will appreciate that the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in the present disclosure. In particular, the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways without departing from the spirit and teachings of the present disclosure. All such combinations and / or couplings fall within the scope of the present disclosure.

[0149] The above describes the embodiments of the present disclosure. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be advantageously used in combination. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A sepsis prediction model training method, characterized in that: include: Acquiring sample time series indicator data, the sample time series indicator data including sample target time series indicator data related to sepsis patients and sample baseline time series indicator data related to non-sepsis patients, the sample time series indicator data including a plurality of sample time series indicators; Dividing the sample time series indicator data into multiple segments of sample cumulative time series indicator data, at least two segments of the sample cumulative time series indicator data have overlapping time ranges; Accumulate time series indicator data for each sample segment and obtain sample statistical characteristics of the sample time series indicator; Determining N sample target time series indicators from the plurality of sample time series indicators based on kernel principal component analysis, where N is a positive integer greater than 0; The sample statistical characteristics and N sample target time series indicators are used to train an initial prediction model to obtain a target prediction model.

2. The training method according to claim 1, characterized in that The step of dividing the sample time series indicator data into multiple segments of sample accumulated time series indicator data includes: Determine the time length and number of segments of the sample time series indicator data; The time length and the number of divided segments are processed based on a division rule to obtain multiple segments of sample accumulated time series indicator data, and the initial moments of the multiple segments of sample accumulated time series indicator data are the same.

3. The training method according to claim 1, characterized in that The sample time series indicators include static indicators and dynamic indicators. The static indicators include age, and the dynamic indicators include systolic blood pressure, diastolic blood pressure, pH value, blood oxygen saturation, body temperature, heart rate, arterial carbon dioxide partial pressure, respiratory rate and white blood cell count.

4. The training method according to claim 3, characterized in that The initial prediction model includes a base learner and a meta learner, The method of training an initial prediction model using the sample statistical features and the sample target time series indicators to obtain a target prediction model includes: Determine the number and type of base learners; For each segment of the sample accumulated time series indicator data, the sample statistical features corresponding to the dynamic indicators, the static features corresponding to the static indicators, and the sample target time series indicators are input into each base learner, and the sample prediction probability is output; Determine a sample prediction matrix based on the plurality of sample prediction probabilities, wherein each row of the prediction matrix represents the prediction probability obtained by all base learners trained based on the same sample accumulated time series data, and each column of the prediction matrix represents the prediction probability obtained by the same base learner trained based on different sample accumulated time series data; Processing the sample prediction matrix using the meta-learner to obtain a sample prediction result; The initial prediction model is trained using the sample prediction results and the label data corresponding to the sample prediction results to obtain a target prediction model.

5. The training method according to claim 4, characterized in that The using the meta-learner to process the sample prediction matrix to obtain a sample prediction result includes: Performing horizontal splicing on the sample prediction matrix to obtain a splicing feature, wherein the horizontal splicing representation splices multiple rows of sample prediction probabilities in the sample prediction matrix into at least one row; The meta-learner is used to process the concatenated features to obtain a sample prediction result.

6. The training method according to claim 4, characterized in that The method of training the initial prediction model using the sample prediction results and the label data corresponding to the sample prediction results to obtain a target prediction model includes: Processing the sample prediction result and the label data according to the loss function to obtain a loss function value; The initial prediction model is trained according to the loss function value to obtain a target prediction model.

7. A method for predicting sepsis, characterized in that: include: Acquire timing indicator data of the tester, wherein the timing indicator data includes a plurality of timing indicators; Dividing the time series indicator data into multiple segments of accumulated time series indicator data, at least two segments of the accumulated time series indicator data have overlapping time ranges; For each segment of accumulated time series indicator data, obtaining statistical characteristics of the time series indicator; Determining N target timing indicators from the plurality of timing indicators based on kernel principal component analysis, where N is a positive integer greater than 0; The statistical features and the N target time series indicators are processed using a target prediction model to obtain a prediction result, wherein the target prediction model is obtained according to the training method described in any one of claims 1 to 6.

8. A sepsis prediction model training device, characterized in that: include: a sample acquisition module, configured to acquire sample time series indicator data, wherein the sample time series indicator data includes sample target time series indicator data related to sepsis patients and sample baseline time series indicator data related to non-sepsis patients, and the sample time series indicator data includes a plurality of sample time series indicators; a sample division module, configured to divide the sample time series indicator data into multiple segments of sample cumulative time series indicator data, wherein at least two segments of the sample cumulative time series indicator data have overlapping time ranges; A sample statistical feature acquisition module is used to accumulate time series indicator data for each sample segment and obtain the sample statistical features of the sample time series indicator; a sample target timing indicator determination module, configured to determine N sample target timing indicators from the plurality of sample timing indicators based on kernel principal component analysis, where N is a positive integer greater than 0; The training module is used to train an initial prediction model using the sample statistical characteristics and N sample target time series indicators to obtain a target prediction model.

9. A sepsis prediction device, characterized in that: include: An acquisition module, configured to acquire timing indicator data of a tester, wherein the timing indicator data includes a plurality of timing indicators; a division module, configured to divide the time series indicator data into multiple segments of accumulated time series indicator data, wherein at least two segments of the accumulated time series indicator data have overlapping time ranges; A statistical feature acquisition module, configured to acquire statistical features of the time series indicator for each segment of accumulated time series indicator data; a target timing indicator determination module, configured to determine N target timing indicators from the plurality of timing indicators based on kernel principal component analysis, where N is a positive integer greater than 0; The prediction result obtaining module is used to process the statistical features and the N target time series indicators using the target prediction model to obtain the prediction result.

10. An electronic device comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

Images