Method and system for predicting septicemia risk based on transfer learning and time sequence feature mining

Abstract

The invention relates to the technical field of medical data analysis, in particular to a septicemia risk prediction method and system based on transfer learning and time sequence feature mining, and the method comprises the steps: obtaining a physiological index data set of a patient, and carrying out the filling preprocessing of a missing value based on a time sequence; extracting multi-scale time sequence features of the physiological index data set by using wavelet transform, and screening the multi-scale time sequence features based on feature importance; based on a pre-training time sequence model, optimizing the pre-training time sequence model according to the screened time sequence features by utilizing transfer learning and combining with the target patient group data; and outputting a septicemia risk level prediction result of the target patient based on the optimized pre-training time sequence model. The objective of the invention is to improve the precision and calculation efficiency of septicemia risk level prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data analysis, and in particular to a method and system for predicting sepsis risk based on transfer learning and time series feature mining. Background Art

[0002] Sepsis is a systemic inflammatory response syndrome caused by the invasion of multiple pathogenic microorganisms into the blood circulation, with high morbidity and mortality. Early identification and intervention are the key to improving patient prognosis. If not diagnosed and treated in time, sepsis may cause serious complications such as myocarditis, gastrointestinal bleeding, uremia, and even lead to multiple organ failure. Traditional early identification of sepsis mainly relies on the experience and judgment of clinicians, but this method has the problems of strong subjectivity and high delay, making it difficult to achieve timely and accurate predictions.

[0003] In order to improve the early prediction ability of sepsis, the existing technology has proposed a prediction method based on machine learning. This method constructs a sepsis prediction model by collecting multiple physiological indicator data of sepsis patients at different periods; then collects the physiological indicator data of the target patient in the current period, extracts the target feature data, and evaluates the risk level of sepsis in the future based on the prediction model. This method improves the accuracy of prediction to a certain extent by extracting the feature data of physiological indicators and using machine learning models for prediction. However, the existing methods still have significant limitations in practical applications: first, due to the high dimensionality and time series complexity of physiological indicator data, the existing methods fail to fully explore the multi-scale time series features in the data during the feature extraction process, resulting in insufficient prediction accuracy and limiting the prediction performance of the model; second, the prediction process of the existing technology is time-consuming and difficult to meet the needs of clinical real-time prediction. Summary of the invention

[0004] In order to improve the accuracy and computational efficiency of sepsis risk level prediction, the present invention provides a sepsis risk prediction method and system based on transfer learning and time series feature mining. The technical solutions adopted are as follows: The technical solution of the first aspect of the present invention provides a method for predicting sepsis risk based on transfer learning and time series feature mining, the method comprising: Obtain the patient's physiological indicator data set and perform preprocessing to fill in missing values ​​based on time series; Wavelet transform is used to extract multi-scale time series features of physiological index data sets, and the multi-scale time series features are screened based on feature importance; Based on the pre-trained time series model, transfer learning is used to combine the target patient group data and optimize the pre-trained time series model according to the screened time series features; Based on the optimized pre-trained time series model, the prediction results of the sepsis risk level of the target patient are output.

[0005] Furthermore, the patient's physiological indicator data set is obtained, and the missing values ​​are preprocessed based on the time series, including: Perform missing value detection on the patient's physiological indicator data set and extract the missing time point set; Classify the missing windows according to the missing time point set, and fill the missing values ​​of the physiological indicator data set based on the category of the missing windows; The filled physiological index data are standardized according to individual patients.

[0006] Furthermore, the missing windows are classified according to the missing time point set, and the missing values ​​of the physiological indicator data set are filled based on the categories of the missing windows, including: For continuous missing windows, linear interpolation is used to fill missing values ​​in the physiological indicator data set, which can be expressed as:

[0007] In the formula, Indicates The patient's The indicators at the time point Fill value of Indicates the offset calculated from the start time of the missing window; represents the number of consecutive missing time points; Indicates The patient's The indicators at the time point The effective value of Indicates The patient's The indicators at the time point Valid values.

[0008] Furthermore, the missing windows are classified according to the missing time point set, and the missing values ​​of the physiological indicator data set are filled based on the categories of the missing windows, which also includes: For non-continuous missing windows, adaptive spline interpolation is used to fill missing values ​​in the physiological index data set, which can be expressed as:

[0009] In the formula, Indicates The patient's The indicators at the time point Fill value of Indicates indivual The spline basis function at time point The value of Indicates patient no. The indicator is The spline coefficients on the basis functions; Indicates patient no. Cubic spline interpolation function of the index; express The number of spline basis functions.

[0010] Furthermore, wavelet transform is used to extract multi-scale time series features of the physiological index data set, and the multi-scale time series features are screened based on feature importance, including: Perform discrete wavelet transform on the physiological index time series data set according to the preset decomposition level, and output a multi-scale wavelet coefficient matrix; Calculate the comprehensive fluctuation intensity of the multi-scale wavelet coefficient matrix; The screened multi-scale time series feature set was extracted based on the average value of the comprehensive fluctuation intensity of the patient group.

[0011] Furthermore, the expression for performing discrete wavelet transform on the physiological index time series data set according to the preset decomposition level is:

[0012] In the formula, Indicates The patient's The indicator is Layer wavelet decomposition time point The wavelet coefficients of ; Indicates that the wavelet basis function is Layer time point For the original time point The response value, Represents the translation index of the wavelet basis function; Indicates The patient's The indicators are at the original time point The standardized value of Indicates the total number of time points.

[0013] Furthermore, based on the pre-trained time series model, transfer learning is used in combination with the target patient group data to optimize the pre-trained time series model according to the screened time series features, including: Align the feature distribution of the source domain and the target domain based on maximizing the difference minimization; The pre-trained time series model is adjusted in combination with the cross entropy loss function, and the parameters are optimized using the target domain data.

[0014] Furthermore, the expression for aligning the feature distribution of the source domain and the target domain based on maximizing the difference minimization is:

[0015] In the formula, represents the maximum mean difference loss; represents the number of patients in the source domain; represents the number of patients in the target domain; represents the kernel mapping function; Indicates the source domain Temporal characteristics of each patient after screening; Indicates the target domain Temporal characteristics of each patient after screening; represents the reproducing kernel Hilbert space; represents the kernel space norm.

[0016] Furthermore, based on the optimized pre-trained time series model, the prediction results of the sepsis risk level of the target patient are output, including: The time series features of the screened target patients are input into the optimized pre-trained time series model, and the sepsis risk probability prediction results of each target patient are output; According to the sepsis risk probability prediction results, the target patients are divided into different risk levels.

[0017] The technical solution of the second aspect of the present invention provides a sepsis risk prediction system based on transfer learning and time series feature mining, using the sepsis risk prediction method based on transfer learning and time series feature mining described in the technical solution of the first aspect of the present invention, the system comprising: A data acquisition and preprocessing module, configured to acquire a patient's physiological indicator data set and perform preprocessing to fill in missing values ​​based on a time series; a feature extraction module configured to extract multi-scale time series features of a physiological indicator data set using wavelet transform, and to screen the multi-scale time series features based on feature importance; A transfer learning module is configured to optimize the pre-trained time series model based on the pre-trained time series model by combining the target patient population data with transfer learning according to the screened time series features; The sepsis prediction module is configured to output a prediction result of the sepsis risk level of the target patient based on the optimized pre-trained time series model.

[0018] The present invention has the following beneficial effects: The sepsis risk prediction method and system based on transfer learning and time series feature mining provided by the present invention utilizes wavelet transform to extract multi-scale time series features of physiological indicators and screens the multi-scale time series features based on feature importance, which can capture local and global dynamic change patterns, screen out features that are highly correlated with sepsis risk, and fully reflect the patient's physiological state while reducing redundant features and reducing the computational complexity of the model; finally, transfer learning is applied based on the pre-trained time series model, combined with the target patient group data and the screened time series features, and the source domain and target domain feature distributions are aligned by maximizing the difference minimization, and the model is optimized in combination with the cross entropy loss function, so that the model can better adapt to the target patient group, improve the prediction accuracy and generalization ability, and solve the performance bottleneck of the prediction model under limited data conditions. This method achieves high-precision and high-efficiency prediction of sepsis risk level. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A method flow chart of a sepsis risk prediction method based on transfer learning and time series feature mining provided by one embodiment of the present invention; Figure 2 A schematic diagram of the structure of a sepsis risk prediction system based on transfer learning and time series feature mining provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the method and system for predicting sepsis risk based on transfer learning and time series feature mining proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0022] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0023] The following is a detailed description of a method and system for predicting sepsis risk based on transfer learning and time series feature mining provided by the present invention in conjunction with the accompanying drawings.

[0024] See also Figure 1 , which shows a method flow chart of a sepsis risk prediction method based on transfer learning and time series feature mining provided by an embodiment of the present invention, the method comprising: Step S100: Acquire the patient's physiological indicator data set, and fill in the missing values ​​based on the time series. Specifically, this embodiment acquires the physiological indicator data set of all sepsis patients in the past three years from the hospital's information system. The physiological indicator data at least includes: body temperature (℃), systolic blood pressure (mmHg), diastolic blood pressure (mmHg), heart rate (times / minute), respiratory rate (times / minute), blood oxygen saturation (%) and other indicators; the patient group set can be expressed as , Represents the number of patients; the time series of physiological indicators of each patient can be expressed as: , Indicates the total number of time points; Indicates Patients in Moment Physiological indicators, , including body temperature, systolic blood pressure, diastolic blood pressure, heart rate, respiratory rate, and blood oxygen saturation; Step S100 specifically includes: Step S110: performing missing value detection on the patient's physiological indicator data set to extract the missing time point set; Step S120: Classify the missing windows according to the missing time point set, and fill the missing values ​​of the physiological indicator data set based on the category of the missing window; specifically, for continuous missing values, use the sliding window technology, if multiple missing values ​​appear continuously within a certain time range, it is determined to be a continuous missing window; for non-continuous missing values, form a non-continuous missing time point set by recording the time points of the missing values; For the continuous missing windows, linear interpolation is used to fill the missing values ​​of the physiological index data set, and linear interpolation is performed using the previous and next valid data points, which can be expressed as:

[0025] In the formula, Indicates The patient's The indicators at the time point Fill value of Indicates the offset calculated from the start time of the missing window; Indicates The patient's The indicators at the time point The effective value of Indicates The patient's The indicators at the time point The effective value of represents the number of consecutive missing time points, , that is, fill in a maximum of 5 hours of continuous missing; For the non-continuous missing time points of the patient, this embodiment takes into account the overall trend of the physiological indicator time series, fits the cubic spline function based on the patient's complete physiological indicator time series, and the filling expression is:

[0026] In the formula, Indicates The patient's The indicators at the time point Fill value of Indicates indivual The spline basis function at time point The value of Indicates patient no. The indicator is The spline coefficients on the basis functions; Indicates patient no. Cubic spline interpolation function of the index; express The number of spline basis functions. The spline coefficients of the patient can be solved by minimizing the objective function, which is expressed as:

[0027] In the formula, Indicates The set of observation time points of the patient, that is, the set of time points of all non-missing values; represents the smoothing parameter; Indicates The patient's The indicators at the time point Fill value of Indicates the maximum value of the time range; Indicates the minimum value of the time range; Represents a spline function The second derivative of Step S130: Standardize the filled physiological indicator data according to individual patients; specifically, in actual medical scenarios, different patients have different basic physiological states. In order to eliminate the impact of such differences on subsequent analysis, this embodiment calculates the mean and standard deviation of the filled physiological indicator data according to individual patients, completes the standardization process, and finally outputs the preprocessed data set, which can be expressed as , , ; It represents the preprocessed data set, which contains the physiological index data of all patients after standardized processing; Indicates The sub-dataset of a patient contains all the time points of the patient of The physiological index data after standardization; Indicates Patients at time point of Physiological index vector after dimension standardization; This is the first Standardized values ​​of physiological indicators.

[0028] This embodiment aims to improve data quality through a systematic data acquisition and preprocessing process. In actual medical application scenarios, the integrity of the patient's physiological indicator data is ensured, and the missing value filling strategy provided by this embodiment avoids information loss and deviation caused by missing data. Standardization processing eliminates the interference of differences in the basic physiological state of different patients on the analysis, so that the data of different patients are on the same comparable scale, and enhances the consistency and availability of the data. Not only does it improve the reliability of the data, but it also improves the accuracy and stability of subsequent feature extraction and model training, which helps to more accurately predict the risk of sepsis and provide clinicians with a more reliable basis for decision-making.

[0029] Step S200: extracting multi-scale time series features of the physiological indicator data set using wavelet transform, and screening the multi-scale time series features based on feature importance; specifically, this embodiment considers the physiological change characteristics at different time scales during the development of sepsis; for example, in the early stage of sepsis, some short-term physiological fluctuations may be signals of disease onset; as the disease progresses, medium-term trends and long-term trends will appear, so this hierarchical decomposition can comprehensively capture the changes in physiological characteristics at different stages.

[0030] Step S200 specifically includes: Step S210: Perform discrete wavelet transform on the physiological index time series data set according to the preset decomposition layer number, and output a multi-scale wavelet coefficient matrix; specifically, the wavelet basis function of this embodiment is preferably the Daubechies wavelet basis, and the decomposition layer number is preferably 3 layers, the purpose of which is to decompose the physiological index time series data into signals of different time scales through discrete wavelet transform, and the 3-layer decomposition layer number corresponds to the following time scales: The first layer is used to capture short-term fluctuations of 0.5-1 hour, such as sudden changes in heart rate; The second layer is used to capture medium-term trends of 1-4 hours, such as a slow drop in blood; The third layer is used to capture long-term changes of 4-24 hours, such as a continuous increase in body temperature; The discrete wavelet transform of the patient's physiological index time series data can be expressed as:

[0031] In the formula, Indicates The patient's The indicator is Layer wavelet decomposition time point The wavelet coefficients of ; Indicates that the wavelet basis function is Layer time point For the original time point The response value, Represents the translation index of the wavelet basis function; Indicates The patient's The indicators are at the original time point The standardized value of Indicates the total number of time points; Finally, the multi-scale wavelet coefficient matrix can be obtained , It represents the dimension of physiological indicators, and 3 represents the number of decomposition levels.

[0032] Step S220: Calculate the comprehensive fluctuation intensity of the multi-scale wavelet coefficient matrix; Specifically, this embodiment quantifies the dynamic abnormality of the physiological index by fusing variance and energy, and the variance can be expressed as:

[0033]

[0034] In the formula, Indicates The patient's The indicator is Variance of the layer; Indicates The patient's The indicator is The variance reflects the degree of dispersion of the wavelet coefficients of the patient's indicator in the corresponding layer, and reflects the fluctuation stability of the indicator on the corresponding time scale. For example, when analyzing the heart rate indicator, a large variance may mean that the heart rate fluctuates violently within the time scale.

[0035] Energy can be expressed as:

[0036] In the formula, Indicates The patient's The indicator is The energy of the layer; the energy reflects the signal strength of the indicator on the corresponding time scale. The higher the energy, the more significant the signal on the time scale.

[0037] The comprehensive volatility intensity can be expressed as:

[0038] In the formula, Indicates The patient's The indicator is The comprehensive fluctuation intensity under the wavelet decomposition of the layer; the comprehensive fluctuation intensity combines the variance and energy indicators to more comprehensively quantify the dynamic abnormality of physiological indicators. For example, for blood pressure indicators, a high comprehensive fluctuation intensity may indicate that blood pressure has both large fluctuations and strong signal changes within the corresponding time scale, which is more likely to be associated with the risk of sepsis.

[0039] Step S230: extract the screened multi-scale time series feature set according to the average comprehensive fluctuation intensity of the patient group; specifically, this implementation screens the features highly correlated with the risk of sepsis through the average comprehensive fluctuation intensity value of the patient group to avoid manual parameter adjustment; the quantile threshold preferably retains the top 20% of the high comprehensive fluctuation intensity features, adaptively matches the data distribution, and the group average comprehensive fluctuation intensity can be expressed as:

[0040] In the formula, Represents the patient population The indicator is The average value of the comprehensive fluctuation intensity under the layer wavelet decomposition, Represents the total number of patients; by calculating the average comprehensive fluctuation intensity of the patient group's item index at the layer, we can understand the fluctuation of the entire patient group on different indicators and time scales. The top 20% of high comprehensive fluctuation intensity features are retained because these features show large fluctuations in the group and are more likely to be highly correlated with sepsis risk. This screening method based on group statistics avoids the subjectivity of manual parameter adjustment and can adaptively match data distribution. The time series features after selection can be expressed as: , To retain the number of features; This embodiment uses multi-scale feature extraction and screening, and discrete wavelet transform decomposes the physiological indicator time series data according to different time scales, which can comprehensively capture the changes in physiological characteristics at each stage of sepsis development, from early short-term fluctuations to medium-term and long-term trends, providing rich information for prediction. Then, by calculating the comprehensive fluctuation intensity, integrating variance and energy, the dynamic abnormalities of physiological indicators can be more accurately quantified, which helps to mine potential risk signals. This embodiment is based on feature screening based on patient group statistics, avoiding the subjectivity of manual parameter adjustment, adaptively matching data distribution, screening out features that are highly correlated with sepsis risk, and reducing redundancy. It not only reduces the complexity of subsequent model calculations, but also improves the model prediction accuracy and generalization ability, making the prediction model more stable and accurate in different patient groups and medical environments.

[0041] Step S300: Based on the pre-trained time series model, transfer learning is used to combine the target patient group data, and the pre-trained time series model is optimized according to the screened time series features; specifically, the source domain data of this embodiment is: multi-scale time series features of historical patients And tags ; The target domain data is the post-screening time series characteristics of the target patient population ; The LSTM model is preferred as the pre-trained time series model; Step S300 specifically includes: Step S300 specifically includes: Step S310: Aligning the source domain and target domain feature distribution based on maximizing the difference minimization, which can be expressed as:

[0042] In the formula, represents the maximum mean difference loss; represents the number of patients in the source domain, that is, the number of samples of historical patients; Represents the number of patients in the target domain, which is the number of samples of the target patient group that currently needs to be predicted; represents the kernel mapping function; Indicates the source domain Temporal characteristics of each patient after screening; Indicates the target domain Temporal characteristics of each patient after screening; represents the reproducing kernel Hilbert space; the kernel mapping function is preferably a Gaussian kernel function; represents the kernel space norm.

[0043] Step S320: The pre-trained time series model is adjusted in combination with the cross entropy loss function, and the parameters are optimized using the target domain data, which can be expressed as:

[0044] In the formula, Indicates the target domain The true sepsis risk signature for each patient Indicates the pre-trained model for the target domain The predicted value of the true sepsis risk probability for each patient; The optimization goal is:

[0045] In the formula, represents the optimized model parameters; Represents a hyperparameter; during the optimization process, the Adam optimizer is used to update the model parameters.

[0046] In actual medical scenarios, this embodiment aligns the feature distribution of the source domain and the target domain by the maximum mean difference, effectively overcoming the problem of inconsistent data distribution in the source domain and the target domain. This enables the model to fully utilize the knowledge contained in the rich historical patient data in the source domain and successfully migrate it to the target patient group, greatly enhancing the generalization ability of the model between different patient groups. At the same time, the model is optimized in combination with the cross entropy loss function to ensure the prediction accuracy of the model in the target domain and significantly improve the accuracy of sepsis risk prediction. It provides more timely and accurate support for the early diagnosis and treatment of sepsis, helping clinicians to more accurately judge patients' sepsis risks and develop more effective treatment plans.

[0047] Step S400: Based on the optimized pre-trained time series model, the prediction result of the sepsis risk level of the target patient is output; in actual clinical applications, the multi-scale time series feature data after the target patient group is screened is obtained from the hospital's electronic medical record system or data storage platform. These data have been extracted and screened in step S200 in the early stage and have features highly correlated with the sepsis risk; then the pre-trained time series model optimized in step S300 is loaded for prediction; specifically, it includes: Step S410: Input the multi-scale time series features after the target patient screening into the optimized pre-trained time series model. The model will output the sepsis risk probability of each target patient, which can be expressed as:

[0048] In the formula, represents the optimized time series model; the risk probability prediction values ​​of all target patients can be composed of a length of Vector , Indicates the target domain The probability of sepsis risk for each patient; Step S420: According to the predicted sepsis risk probability, the target patient is divided into different risk levels, which can be expressed as:

[0049] Ultimately, a corresponding treatment plan can be formulated according to the risk level. For example, for low-risk patients, follow the established medical process and appropriately reduce unnecessary monitoring frequency, such as adjusting the monitoring of physiological indicators from once an hour to once every 4 hours. At the same time, reduce the intensity of treatment to avoid excessive medical treatment and save medical resources. For example, for medium-risk patients, strengthen monitoring efforts, increase monitoring items or increase monitoring frequency, such as increasing the frequency of blood tests, and take preventive measures, such as using antibiotics in advance to prevent worsening of infection. For example, for high-risk patients, initiate emergency treatment plans, organize multidisciplinary teams for consultations, and develop comprehensive and proactive treatment intervention plans; This embodiment ensures the high accuracy and reliability of sepsis risk probability prediction by inputting the characteristics of the target patient into the pre-trained time series model after transfer learning and optimization of the target domain data. The model has undergone multi-step optimization in the early stage, fully learned the knowledge of the source domain data and adapted to the characteristics of the target patient group. The risk probability is further divided into intuitive risk levels, which greatly facilitates the work of clinicians. Doctors can quickly and clearly understand the sepsis risk level of each patient, and thus formulate highly targeted treatment plans based on the risk level.

[0050] In summary, the sepsis risk prediction method based on transfer learning and time series feature mining provided by the present invention significantly improves the prediction accuracy and the computational efficiency of the prediction process through multi-stage collaborative optimization. First, based on the missing value filling of the time series, based on linear interpolation, adaptive spline interpolation and individualized standardization processing, data noise and individual differences are effectively eliminated to ensure a high signal-to-noise ratio of the input features; then, multi-scale time series features are extracted through wavelet transform, and key features are screened in combination with comprehensive fluctuation intensity, and the short-term fluctuations, medium-term trends and long-term abnormalities of physiological indicators are accurately quantified, thereby enhancing the sensitivity to early signals of sepsis; then, the source domain historical data is used to pre-train the model, and the target domain feature distribution is aligned through the maximum mean difference loss, combined with cross entropy loss fine-tuning, to achieve high generalization performance in data-limited scenarios; the finally optimized lightweight model supports millisecond-level reasoning, and combines dynamic thresholds to divide risk levels (low / medium / high risk), providing clinicians with interpretable and actionable early warning results to facilitate early intervention. This method takes into account prediction accuracy, computational efficiency and clinical interpretability through multi-scale feature fusion, cross-domain knowledge transfer and efficient computing architecture, providing a reliable technical means for the precise prevention and control of sepsis.

[0051] See also Figure 2 , which shows a schematic diagram of the structure of a sepsis risk prediction system based on transfer learning and time series feature mining provided by an embodiment of the present invention, the system comprising: A data acquisition and preprocessing module, configured to acquire a patient's physiological indicator data set and perform preprocessing to fill in missing values ​​based on a time series; a feature extraction module configured to extract multi-scale time series features of a physiological indicator data set using wavelet transform, and to screen the multi-scale time series features based on feature importance; A transfer learning module is configured to optimize the pre-trained time series model based on the pre-trained time series model by combining the target patient population data with transfer learning according to the screened time series features; The sepsis prediction module is configured to output a prediction result of the sepsis risk level of the target patient based on the optimized pre-trained time series model.

[0052] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0053] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A sepsis risk prediction method based on transfer learning and time series feature mining, characterized in that: The method comprises: Obtain the patient's physiological indicator data set and perform preprocessing to fill in missing values ​​based on time series; Wavelet transform is used to extract multi-scale time series features of physiological index data sets, and the multi-scale time series features are screened based on feature importance; Based on the pre-trained time series model, transfer learning is used to combine the target patient group data and optimize the pre-trained time series model according to the screened time series features; Based on the optimized pre-trained time series model, the prediction results of the sepsis risk level of the target patient are output.

2. The method for predicting sepsis risk according to claim 1, wherein: Obtain the patient's physiological indicator data set and perform preprocessing to fill in missing values ​​based on the time series, including: Perform missing value detection on the patient's physiological indicator data set and extract the missing time point set; Classify the missing windows according to the missing time point set, and fill the missing values ​​of the physiological indicator data set based on the category of the missing windows; The filled physiological index data are standardized according to individual patients.

3. The method for predicting sepsis risk according to claim 2, wherein: The missing windows are classified according to the missing time point set, and the missing values ​​of the physiological indicator dataset are filled based on the category of the missing windows, including: For continuous missing windows, linear interpolation is used to fill missing values ​​in the physiological indicator data set, which can be expressed as: In the formula, Indicates The patient's The indicators at the time point Fill value of Indicates the offset calculated from the start time of the missing window; represents the number of consecutive missing time points; Indicates The patient's The indicators at the time point The effective value of Indicates The patient's The indicators at the time point Valid values.

4. The method for predicting sepsis risk according to claim 3, wherein: The missing windows are classified according to the missing time point set, and the missing values ​​of the physiological indicator data set are filled based on the category of the missing windows, and also include: For non-continuous missing windows, adaptive spline interpolation is used to fill missing values ​​in the physiological index data set, which can be expressed as: In the formula, Indicates The patient's The indicators at the time point Fill value of Indicates indivual The spline basis function at time point The value of Indicates patient no. The indicator is The spline coefficients on the basis functions; Indicates patient no. Cubic spline interpolation function of the index; express The number of spline basis functions.

5. The method for predicting sepsis risk according to claim 1, wherein: Wavelet transform is used to extract multi-scale time series features of physiological indicator data sets, and multi-scale time series features are screened based on feature importance, including: Perform discrete wavelet transform on the physiological index time series data set according to the preset decomposition level, and output a multi-scale wavelet coefficient matrix; Calculate the comprehensive fluctuation intensity of the multi-scale wavelet coefficient matrix; The screened multi-scale time series feature set was extracted based on the average value of the comprehensive fluctuation intensity of the patient group.

6. The method for predicting sepsis risk according to claim 5, wherein: The expression for performing discrete wavelet transform on the physiological index time series data set according to the preset decomposition level is: In the formula, Indicates The patient's The indicator is Layer wavelet decomposition time point The wavelet coefficients of ; Indicates that the wavelet basis function is Layer time point For the original time point The response value, Represents the translation index of the wavelet basis function; Indicates The patient's The indicators are at the original time point The standardized value of Indicates the total number of time points.

7. The method for predicting sepsis risk according to any one of claims 1 to 6, characterized in that: Based on the pre-trained time series model, transfer learning is used to combine the target patient group data and optimize the pre-trained time series model according to the screened time series features, including: Align the feature distribution of the source domain and the target domain based on maximizing the difference minimization; The pre-trained time series model is adjusted in combination with the cross entropy loss function, and the parameters are optimized using the target domain data.

8. The method for predicting sepsis risk according to claim 7, wherein: The expression for aligning the feature distribution of the source domain and the target domain based on maximizing the difference minimization is: In the formula, represents the maximum mean difference loss; represents the number of patients in the source domain; represents the number of patients in the target domain; represents the kernel mapping function; Indicates the source domain Temporal characteristics of each patient after screening; Indicates the target domain Temporal characteristics of each patient after screening; represents the reproducing kernel Hilbert space; represents the kernel space norm.

9. The method for predicting sepsis risk according to claim 7, wherein: Based on the optimized pre-trained time series model, the sepsis risk level prediction results of the target patient are output, including: The time series features of the screened target patients are input into the optimized pre-trained time series model, and the sepsis risk probability prediction results of each target patient are output; According to the sepsis risk probability prediction results, the target patients are divided into different risk levels.

10. A sepsis risk prediction system based on transfer learning and time series feature mining, characterized in that: The method for predicting sepsis risk based on transfer learning and time series feature mining according to any one of claims 1 to 9 is adopted, wherein the system comprises: A data acquisition and preprocessing module, configured to acquire a patient's physiological indicator data set and perform preprocessing to fill in missing values ​​based on a time series; a feature extraction module configured to extract multi-scale time series features of a physiological indicator data set using wavelet transform, and to screen the multi-scale time series features based on feature importance; A transfer learning module is configured to optimize the pre-trained time series model based on the pre-trained time series model by combining the target patient population data with transfer learning according to the screened time series features; The sepsis prediction module is configured to output a prediction result of the sepsis risk level of the target patient based on the optimized pre-trained time series model.

Images