Sepsis prediction method and system based on machine learning

Abstract

The present invention relates to the field of medical data processing technology, and in particular to a sepsis prediction method and system based on machine learning, including: obtaining multiple one-dimensional data sets, recording multiple data of a patient at the same time as a data sequence; obtaining a sepsis performance factor of each dimensional data according to the value of each data in the one-dimensional data set of each dimension; combining the data difference between each data sequence and other data sequences in the same dimension, obtaining the distance between each data sequence and other data sequences, obtaining the clustering radius of each data sequence, thereby obtaining a quantity threshold of the data sequence, clustering the data sequence into multiple clusters, and obtaining a sepsis prediction model. The present invention aims to solve the problem that the DBSCAN algorithm cannot filter out data showing sepsis characteristics from the ICU data set, thereby affecting the recognition effect of early signs of sepsis.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and in particular to a sepsis prediction method and system based on machine learning. Background Art

[0002] Sepsis prediction methods have been an important research direction in the medical field in recent years, aiming to achieve timely and effective treatment and reduce the mortality rate of sepsis by early identification and prediction of sepsis. In view of the high incidence and poor prognosis of sepsis, there are already a variety of models that combine machine learning to assess the risk of patients in intensive care units. These models help identify early signs of sepsis by analyzing a large amount of clinical data, thereby achieving prediction. However, in the process of training the model, due to the current lack of sepsis-related datasets, it is necessary to screen relevant data in other ICU datasets.

[0003] When the current DBSCAN clustering algorithm is used to screen out sepsis-related data sets from the ICU data set, the same clustering radius is used for each data, and the quantity threshold is preset; during the clustering process, data showing sepsis characteristics and data not showing sepsis characteristics are clustered into one category, making the model based on the clustering results inaccurate; resulting in the sepsis prediction model constructed by machine learning being unable to help identify early signs of sepsis. Summary of the invention

[0004] The present invention provides a sepsis prediction method and system based on machine learning to solve the existing problems.

[0005] The sepsis prediction method and system based on machine learning of the present invention adopts the following technical solutions:

[0006] The present invention proposes a sepsis prediction method based on machine learning, which comprises the following steps:

[0007] According to the ICU medical data set, a one-dimensional data set of multiple dimensions and a critical value of each dimension are obtained to classify the dimensions; the data of multiple dimensions obtained from a patient at the same acquisition time are recorded as a data sequence; multiple data sequences generated by a patient during a treatment process are recorded as data sequences during the same treatment process; in the one-dimensional data set, each data corresponds to an acquisition time;

[0008] Fit all data of each dimension in the data sequence of the same treatment process to obtain the fitting slope of each data in the one-dimensional data set of each dimension; obtain the abnormal coefficient of each data in the one-dimensional data set of each dimension according to the critical value of each dimension, the fitting slope and the value of each data in the one-dimensional data set of each dimension; obtain the sepsis manifestation factor of each dimensional data according to the value and abnormal coefficient of each data in the one-dimensional data set of each dimension;

[0009] According to the sepsis manifestation factor of each dimension data and the data difference between each data sequence and other data sequences in the same dimension, the distance between each data sequence and other data sequences is obtained;

[0010] The clustering radius of each data sequence is obtained according to the distance between each data sequence and other data sequences during each treatment process, the sepsis manifestation factor of each dimensional data, and the abnormal coefficient of each data in each data sequence;

[0011] According to the clustering radius of each data sequence and the distance between each data sequence and other data sequences, a threshold value of the number of data sequences is obtained, and all data sequences of all treatment processes are clustered into multiple clusters;

[0012] Different clusters were analyzed, and data sequences for training the sepsis prediction model were screened out to obtain the sepsis prediction model.

[0013] Furthermore, the one-dimensional data sets of multiple dimensions and the critical value of each dimension are obtained according to the ICU medical data set, and the dimensions are classified, including the specific method of:

[0014] Get the dimension data present in the Sepsis-3 standard and get One-dimensional dataset with dimensions Indicates the number of dimensions contained in the Sepsis-3 standard;

[0015] The Sepsis-3 standard has different standard ranges for different dimensions. The data of the dimension is smaller than When the standard range of the dimension is at the lower limit, the patient is infected with sepsis, and the The first dimension is recorded as the first dimension, The lower limit of the standard range of the dimension is recorded as The minimum critical value of the dimension;

[0016] When The data of the dimension is greater than When the upper limit of the standard range of the first dimension is reached, the patient is infected with sepsis. The dimension is recorded as the second dimension, and the The upper limit of the standard range of the dimension is recorded as The maximum critical value of the dimension;

[0017] When The data of the dimension is greater than The upper limit of the standard range of the dimension or less than When the standard range of the dimension is at the lower limit, the patient is infected with sepsis, and the dimensions, recorded as the third dimension, The upper limit of the standard range of the dimension is recorded as The maximum critical value of the dimension The lower limit of the standard range of the dimension is recorded as The minimum critical value of each dimension; the maximum critical value and the minimum critical value of each dimension are recorded as the critical value of each dimension.

[0018] Furthermore, the method of fitting all the data in each dimension in the data sequence of the same treatment process to obtain the fitting slope of each data in the one-dimensional data set of each dimension includes the following specific methods:

[0019] The first In a one-dimensional data set of dimensions All the treatments in which the data are located The data in each dimension is mapped into a two-dimensional coordinate system with the horizontal axis being the acquisition time and the vertical axis being the data value;

[0020] The data points in the two-dimensional coordinate system are fitted by the least square method to obtain the In a one-dimensional data set of dimensions All the treatments in which the data are located The fitting curve of the data in the dimension is recorded as In the dimension A fitting curve;

[0021] The first In a one-dimensional data set of dimensions The data in In the dimension The fitting slope of the corresponding data point on the fitting curve is recorded as In a one-dimensional data set of dimensions The fitted slope of the data.

[0022] Furthermore, the abnormal coefficient of each data in the one-dimensional data set of each dimension is obtained according to the critical value of each dimension, the fitting slope and the value of each data in the one-dimensional data set of each dimension, and the specific formula included is:

[0023] When When the dimension is a first-class dimension:

[0024] ;

[0025] When When the dimension is a second-type dimension:

[0026] ;

[0027] When When the dimension is the third dimension:

[0028] ;

[0029] In the formula, Indicates In a one-dimensional data set of dimensions The abnormal coefficient of the data, Indicates In a one-dimensional data set of dimensions The value of the data, Indicates The minimum critical value of the dimension, Indicates In a one-dimensional data set of dimensions The fitted slope of the data, Indicates The maximum critical value of the dimension, represents the sigmoid function, represents the hyperbolic tangent function.

[0030] Furthermore, the method of obtaining the sepsis manifestation factor of each dimensional data according to the value and abnormal coefficient of each data in the one-dimensional data set of each dimension includes the following specific methods:

[0031] The first In a one-dimensional data set of dimensions The value of the data is The square of the difference between the means of all data in the one-dimensional data set of dimensions is recorded as In the dimension The difference factor of each data;

[0032] The first In the dimension The difference factor of the data is In a one-dimensional data set of dimensions The product of the abnormal coefficients of the data is recorded as In the dimension The performance factor of each data;

[0033] The first The normalized value of the mean of the performance factors of all data in the dimension is recorded as Sepsis manifestation factors based on dimensional data.

[0034] Furthermore, the distance between each data sequence and other data sequences is obtained according to the sepsis manifestation factor of each dimensional data and the data difference between each data sequence and other data sequences in the same dimension, including the specific method of:

[0035] ;

[0036] In the formula, Indicates The data series and The distance between the data sequences, Indicates the number of dimensions contained in the Sepsis-3 standard. Indicates In the data series The value of the dimension data, Indicates In the data series The value of the dimension data, Indicates Sepsis manifestation factors based on dimensional data.

[0037] Furthermore, the clustering radius of each data sequence is obtained according to the distance between each data sequence and other data sequences in each treatment process, the sepsis manifestation factor of each dimensional data, and the abnormal coefficient of each data in each data sequence, including the specific method of:

[0038] Calculate the The data series and The difference in the acquisition time of each data series during the treatment process in which the data series are located will be The treatment process in which the data series is located is The data sequence with the largest difference in the collection time among the data sequences is recorded as The corresponding data sequence of a data sequence;

[0039] No. The cluster radius of the data series The specific calculation formula is as follows:

[0040] ;

[0041] In the formula, No. The clustering radius of the data series, Indicates The distance between a data sequence and its corresponding data sequence, Indicates Sepsis manifestation factors of dimensional data, Indicates the number of dimensions contained in the Sepsis-3 standard. Indicates In the data series The abnormal coefficient of the dimensional data.

[0042] Furthermore, the method of obtaining the quantity threshold of the data sequences according to the clustering radius of each data sequence and the distance between each data sequence and other data sequences includes the following specific methods:

[0043] ;

[0044] In the formula, represents the quantity threshold of the data sequence, Indicates the number of data contained in each one-dimensional data set. Indicates The number of data sequences contained within the clustering radius of a data sequence, Indicates The clustering radius of the data series, Indicates The cluster radius of the data series The clustering radius of the data series, Indicates The data sequence and the The distance between the data sequences, is an exponential function with a natural constant as base, Represents the normalization function.

[0045] Furthermore, the specific method of analyzing different clusters and selecting data sequences for training a sepsis prediction model includes:

[0046] The mean of the anomaly coefficients of all data in any data sequence is recorded as the local anomaly factor of the data sequence, and the mean of the local anomaly factors of all data sequences in any cluster is recorded as the average anomaly degree of the cluster;

[0047] According to the average abnormality of each cluster, the clusters used to train the sepsis prediction model are screened out. The screening process is as follows:

[0048] ;

[0049] In the formula, Indicates the number of clusters in the clustering results. represents the number of selected clusters, Indicates the clustering result The average abnormality of the clusters, Indicates the preset selection quantity threshold; when selecting clusters, select them in descending order of average abnormality;

[0050] when The first time greater than or equal to , the selected clusters are used as the filtered data.

[0051] The present invention also proposes a sepsis prediction system based on machine learning, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the steps of the aforementioned sepsis prediction method based on machine learning.

[0052] The technical solution of the present invention has the following beneficial effects: the present invention obtains one-dimensional data sets of multiple dimensions by analyzing multi-dimensional data in a sepsis-related ICU data set; when calculating the abnormal coefficient of each data in the one-dimensional data set of each dimension, not only the difference between each data and its corresponding critical value is used, but also the development trend of each data toward the critical value is combined to ensure the prediction ability of the training model; according to the characteristic that the greater the fluctuation of the data in the one-dimensional data set of each dimension, the more obvious the difference between infected sepsis and non-infected sepsis in this dimension, the sepsis expression factor of each dimensional data is obtained, and a foundation is laid for obtaining the distance between each data sequence and other data sequences; according to the expression degree of sepsis characteristics of the data sequence, the clustering radius of the data sequence is obtained, so that the clustering radius of the data sequence expressing sepsis characteristics is greater than the clustering radius of the data sequence not expressing sepsis characteristics, so that in the clustering process, the data sequence expressing sepsis characteristics is clustered into one category, and the data sequence not expressing sepsis characteristics is clustered into one category, so that the data sequence screened out according to the clustering result is more reliable, and the accuracy of the prediction result of the sepsis prediction model is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be 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.

[0054] Figure 1 The figure is a flowchart of the steps of the sepsis prediction method based on machine learning of the present invention. DETAILED DESCRIPTION

[0055] 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 sepsis prediction method and system based on machine learning proposed by the present invention, its specific implementation, 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.

[0056] 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.

[0057] The specific scheme of the sepsis prediction method and system based on machine learning provided by the present invention is described in detail below with reference to the accompanying drawings.

[0058] See also Figure 1 , which shows a flowchart of a method for predicting sepsis based on machine learning provided by an embodiment of the present invention, the method comprising the following steps:

[0059] Step S001: Using the Sepsis-3 standard and the ICU medical data set, obtain one-dimensional data sets of multiple dimensions and the critical value of each dimension, and classify the dimensions; record the data of multiple dimensions acquired from a patient at the same acquisition time as a data sequence; record multiple data sequences generated by a patient during a treatment process as data sequences during the same treatment process.

[0060] It should be noted that in the process of training the sepsis prediction model, due to the lack of existing data on sepsis, other ICU medical data sets were obtained through the Internet, and the data information in these data sets was selected.

[0061] Specifically, in this embodiment, the dimension data existing in the Sepsis-3 standard in the MIMIC-III database and the PhysioNet / CinC2019 Challenge dataset are obtained to obtain The one-dimensional data set is a one-dimensional data set with dimensions, and the dimension data is the physiological data with time series characteristics of patients in the ICU. Among them, the detection data in the MIMIC-III database and the PhysioNet / CinC2019 Challenge data set include various physiological data with time series characteristics of patients in the ICU, in which the data of different patients and different treatments have corresponding numbers in the dictionary table (subject_id, hadm_id, icustay_id). Indicates the number of dimensions contained in the Sepsis-3 standard.

[0062] Furthermore, the Sepsis-3 standard has different standard ranges for different dimensions. The data of the dimension is smaller than When the standard range of the dimension is at the lower limit, the patient is infected with sepsis, and the The first dimension is recorded as the first dimension, The lower limit of the standard range of the dimension is recorded as The minimum critical value of the dimension; when the The data of the dimension is greater than When the upper limit of the standard range of the first dimension is reached, the patient is infected with sepsis. The dimension is recorded as the second dimension, and the The upper limit of the standard range of the dimension is recorded as The maximum critical value of the dimension; when the The data of the dimension is greater than The upper limit of the standard range of the dimension or less than When the standard range of the dimension is at the lower limit, the patient is infected with sepsis, and the dimensions, recorded as the third dimension, The upper limit of the standard range of the dimension is recorded as The maximum critical value of the dimension The lower limit of the standard range of the dimension is recorded as The minimum critical value of each dimension; the maximum critical value and the minimum critical value of each dimension are recorded as the critical value of each dimension.

[0063] Furthermore, multiple dimensional data acquired from a patient at the same collection time are recorded as a data sequence; and multiple data sequences generated by a patient during a treatment process are recorded as data sequences during the same treatment process.

[0064] Step S002: Fit all data of each dimension in the data sequence of the same treatment process to obtain the fitting slope of each data in the one-dimensional data set of each dimension; obtain the abnormal coefficient of each data in the one-dimensional data set of each dimension according to the critical value of each dimension, the fitting slope and the value of each data in the one-dimensional data set of each dimension; obtain the sepsis manifestation factor of each dimensional data according to the value and abnormal coefficient of each data in the one-dimensional data set of each dimension.

[0065] It should be noted that when obtaining the performance of each dimensional data on sepsis, the abnormal coefficient of each data in the one-dimensional data set of each dimension is first calculated. Then, the sepsis performance factor of each dimensional data is obtained according to the abnormal coefficient and value of each data in the one-dimensional data set of each dimension.

[0066] It should be further explained that, because the difference between each data in each one-dimensional data set and its critical value in each dimension represents the degree of sepsis infection of the patient. Therefore, the difference between each data in each one-dimensional data set and its critical value in each dimension is used as a parameter for calculating the abnormal coefficient of each data in each one-dimensional data set. In addition, in order to ensure the predictive ability of the training model, the one-dimensional data sets of some dimensions may contain normal data that does not exceed the critical value. Therefore, the development trend of each data in each one-dimensional data set to its critical value, that is, the fitting slope of each data in each one-dimensional data set in each dimension, is used as a parameter for calculating the abnormal coefficient of each data in each one-dimensional data set in each dimension. Therefore, according to the difference between the value of each data in each one-dimensional data set in each dimension and its critical value, and the fitting slope of each data in each one-dimensional data set in each dimension, the abnormal coefficient of each data in each one-dimensional data set in each dimension is obtained.

[0067] Specifically, In a one-dimensional data set of dimensions Take the data as an example. In a one-dimensional data set of dimensions All the treatments in which the data are located The data in the first dimension are mapped to a two-dimensional coordinate system with the horizontal axis being the acquisition time and the vertical axis being the data value. The data points in the two-dimensional coordinate system are fitted with the least squares method to obtain the first In a one-dimensional data set of dimensions All the treatments in which the data are located The fitting curve of the data in the dimension is recorded as In the dimension Get the first fitting curve. In the dimension The fitting slope of each data point in the fitting curve is In a one-dimensional data set of dimensions data, in In the dimension The fitting slope of the corresponding data point on the fitting curve is recorded as In a one-dimensional data set of dimensions The fitted slope of the data.

[0068] According to The critical value of the dimension In a one-dimensional data set of dimensions The value of the data and the fitting slope are obtained. In a one-dimensional data set of dimensions The specific calculation formula for the abnormal coefficient of data is as follows:

[0069] When When the dimension is a first-class dimension:

[0070] ;

[0071] When When the dimension is a second-type dimension:

[0072] ;

[0073] When When the dimension is the third dimension:

[0074] ;

[0075] In the formula, Indicates In a one-dimensional data set of dimensions The abnormal coefficient of the data, Indicates In a one-dimensional data set of dimensions The value of the data, Indicates The minimum critical value of the dimension, Indicates In a one-dimensional data set of dimensions The fitted slope of the data, Indicates The maximum critical value of the dimension, represents the sigmoid function, represents the hyperbolic tangent function.

[0076] It should be noted that and Indicates When the dimensions are of different types, In a one-dimensional data set of dimensions The difference between a data and its critical value; When the dimension is a first-class dimension, The smaller the value of In a one-dimensional data set of dimensions The more data there are, the stronger their ability to express sepsis is; When the dimension is a second-type dimension, The larger the value, the In a one-dimensional data set of dimensions The more data there are, the stronger their ability to express sepsis is; When the dimension is the third dimension, The larger the value or The smaller the value of In a one-dimensional data set of dimensions The stronger the performance of the data for sepsis; Indicates In a one-dimensional data set of dimensions Data to the The development trend of the critical value of each dimension When the dimension is a first-class dimension, The smaller the value, the In a one-dimensional data set of dimensions The faster the data develops towards the characteristics of sepsis; When the dimension is a second-type dimension, The larger the value, the In a one-dimensional data set of dimensions The faster the data develops towards the characteristics of sepsis; When the dimension is the third dimension, The smaller or larger the value of In a one-dimensional data set of dimensions The faster the data develop towards sepsis characteristics.

[0077] At this point, the abnormal factor of each data in the one-dimensional data set of each dimension is obtained.

[0078] It should be further explained that for each dimension of the one-dimensional data set, the greater the difference between the data, the wider the distribution range of the data in this dimension, that is, the value of the patient infected with sepsis in this dimension is more obviously different from the value of the patient not infected with sepsis. Therefore, the difference between the value of each data in the one-dimensional data set of each dimension and the mean of all data in the one-dimensional data set of this dimension is used as a parameter for calculating the sepsis performance factor of each dimension of data; because the abnormal factor of each data in the one-dimensional data set of each dimension may be different, the abnormal factor of each data in the one-dimensional data set of each dimension is used as a parameter for calculating the sepsis performance factor of each dimension of data.

[0079] Specifically, the The sepsis performance factor of the dimensional data is recorded as , the specific calculation formula is as follows:

[0080] ;

[0081] In the formula, Indicates Sepsis manifestation factors of dimensional data, Indicates The number of data contained in a one-dimensional data set of dimensions, Indicates In a one-dimensional data set of dimensions The value of the data, Indicates The mean of all data in a one-dimensional data set of dimensions, Indicates In a one-dimensional data set of dimensions The abnormal coefficient of the data, Represents a linear normalization function, whose normalized object is each dimension of data .

[0082] It should be noted that Indicates In the dimension The difference factor of the data reflects the The value of the data is The larger the difference in the mean of all data in the one-dimensional data set of dimensions, the greater the difference in the mean of all data in the one-dimensional data set of dimensions. The greater the possibility that the value of a dimension in which the patient is infected with sepsis is significantly different from that in which the patient is not infected with sepsis; Indicates In a one-dimensional data set of dimensions The larger the abnormal coefficient of the data, the larger the value is. In a one-dimensional data set of dimensions The data and The difference in the mean of all data in the one-dimensional data set of dimensions is The greater the impact of sepsis manifestation factors on the dimensional data; Indicates In the dimension The performance factor of the data.

[0083] At this point, the sepsis manifestation factor of each dimension of data is obtained.

[0084] Step S003: Obtain the distance between each data sequence and other data sequences according to the sepsis manifestation factor of each dimensional data and the data difference between each data sequence and other data sequences in the same dimension.

[0085] It should be noted that, because the sepsis expression factor of each dimensional data is different from the sepsis expression factors of other dimensional data, the distance between each data sequence and other data sequences is obtained according to the sepsis expression factor of each dimensional data and the data difference of different data sequences in the same dimension.

[0086] Specifically, the The data series and The distance between the data sequences is recorded as , the specific calculation formula is as follows:

[0087] ;

[0088] In the formula, Indicates The data series and The distance between the data sequences, Indicates the number of dimensions contained in the Sepsis-3 standard. Indicates In the data series The value of the dimension data, Indicates In the data series The value of the dimension data, Indicates Sepsis manifestation factors based on dimensional data.

[0089] It should be noted that is the Ath data sequence and the The data series is in The larger the value of the numerical difference in the dimensional data, the greater the The data series and The larger the distance between the data sequences, Indicates the Ath data sequence and the The data series is in The numerical difference in the data of the first dimension has an impact on the The data series and The greater the value, the greater the influence of the distance between the Ath data sequence and the The data series is in The numerical difference in the data of the first dimension has an impact on the The data series and The greater the influence of the distance between the two data sequences, the greater the influence of the distance between the two data sequences.

[0090] At this point, the distance between each data sequence and other data sequences is obtained.

[0091] Step S004: Obtain the clustering radius of each data sequence according to the distance between each data sequence and other data sequences during each treatment process, the sepsis manifestation factor of each dimensional data, and the abnormal coefficient of each data in each data sequence.

[0092] It should be noted that the purpose of clustering data sequences is to cluster data sequences that exhibit sepsis characteristics into one category and data sequences that do not exhibit sepsis characteristics into another category. Because the sepsis expression factor of each dimension of data reflects the difference between data that exhibit sepsis characteristics and data that do not exhibit sepsis characteristics in this dimension, the abnormal coefficient of each dimension of data in each data sequence reflects the ability of each dimension of data in this data sequence to express sepsis characteristics; therefore, the abnormal coefficient of each dimension of data in each data sequence and the sepsis expression factor of each dimension of data are used as parameters for calculating the clustering radius of each data sequence.

[0093] It should be further explained that when treating a patient, each data sequence can, to a certain extent, represent the overall health of the patient during the treatment process. Therefore, the distance between each data sequence and other data sequences in the treatment process in which each data sequence is located is used as a parameter for calculating the clustering radius of each data sequence. Thus, the clustering radius of each data sequence is obtained based on the abnormal coefficient of each dimension data in each data sequence, the sepsis manifestation factor of each dimension data, and the overall health of each data sequence and other data sequences in the treatment process in which each data sequence is located.

[0094] Specifically, calculate The data series and The difference in the acquisition time of each data series during the treatment process in which the data series are located will be The treatment process in which the data series is located is The data sequence with the largest difference in the collection time among the data sequences is recorded as The corresponding data sequence of a data sequence.

[0095] Furthermore, The clustering radius of a data sequence is recorded as , the specific calculation formula is:

[0096] ;

[0097] In the formula, No. The clustering radius of the data series, Indicates The distance between a data sequence and its corresponding data sequence, Indicates Sepsis manifestation factors of dimensional data, Indicates the number of dimensions contained in the Sepsis-3 standard. Indicates In the data series The abnormal coefficient of the dimensional data.

[0098] It should be noted that Indicates The data series and The larger the value, the worse the patient's health is during this treatment. The greater the possibility that the patient is infected with sepsis at the collection time corresponding to the first data series, the greater the probability that the patient is infected with sepsis at the collection time corresponding to the first data series. The larger the clustering radius of a data sequence; Indicates The greater the value of the data sequence, the greater the performance of the sepsis information feature. The more obvious the performance of the sepsis information characteristics of the first data sequence is, the The larger the clustering radius of a data sequence, the more treatment data and data with similar information characteristics can be included in its clustering radius, so that its clustering results have a stronger ability to represent the information of sepsis characteristics and provide a reference for the subsequent calculation of the threshold of the number of data points.

[0099] At this point, the clustering radius of each data series is obtained.

[0100] Step S005: According to the clustering radius of each data sequence and the distance between each data sequence and other data sequences, a data sequence quantity threshold is obtained, and all data sequences of all treatment processes are clustered into multiple clusters.

[0101] It should be noted that, because the quantity thresholds of different data sequences are the same, the quantity threshold of data sequences is obtained according to the clustering radius of each data sequence, the distance between each data sequence and all data sequences within its clustering radius, and the number of data sequences contained in the clustering radius of each data sequence.

[0102] Specifically, it will be The distance between the data sequences is less than or equal to the All data sequences with clustering radius of data sequences are recorded as The data sequences within the clustering radius of the data sequences.

[0103] Furthermore, the number threshold of the data sequence is recorded as , the calculation formula is as follows:

[0104] ;

[0105] In the formula, represents the quantity threshold of the data sequence, Indicates the number of data contained in each one-dimensional data set. Indicates The number of data sequences contained within the clustering radius of a data sequence, Indicates The clustering radius of the data series, Indicates The cluster radius of the data series The clustering radius of the data series, Indicates The data sequence and the The distance between the data sequences, is an exponential function with a natural constant as the base. The model shows an inverse proportional relationship. As the input of the model, the implementer can set the inverse proportional function and normalization function according to the actual situation. Represents a linear normalization function, whose normalization object is each data sequence .

[0106] It should be noted that , indicating the The distribution density of other data sequences around the data sequence. The smaller the value, the denser the distribution density. The greater the distribution density of other data sequences around a data sequence.

[0107] At this point, the quantity threshold of the data sequence is obtained.

[0108] According to the clustering radius of each data sequence and the number threshold of the data sequence, the data sequences are clustered into multiple clusters through the DBSCAN algorithm.

[0109] Step S006: Analyze different clusters, screen out data sequences for training a sepsis prediction model, and obtain a sepsis prediction model.

[0110] Specifically, the mean of the anomaly coefficients of all data in any data sequence is recorded as the local anomaly factor of the data sequence, and the mean of the local anomaly factors of all data sequences in any cluster is recorded as the average anomaly degree of the cluster.

[0111] Furthermore, based on the average abnormality of each cluster, the clusters used to train the sepsis prediction model are obtained. The screening process is as follows:

[0112] ;

[0113] In the formula, Indicates the number of clusters in the clustering results. represents the number of selected clusters, Indicates the clustering result The average abnormality of the clusters, Indicates the selection quantity threshold. The selection quantity threshold preset in this embodiment , this is taken as an example for description, and other implementations may be set to other values. When selecting clusters, they are selected in descending order of average abnormality.

[0114] Furthermore, when The first time greater than or equal to When , several clusters selected at this time are used as the number samples after screening.

[0115] After obtaining the filtered data, the existing machine learning model is used to train the sepsis prediction model using the filtered data as training samples. The training process can refer to the title "Real-time Risk Prediction Model of Sepsis in Critically Ill Patients Based on Machine Learning" [J], a journal of the Chinese Journal of Biomedical Engineering.

[0116] At this point, this embodiment is completed.

[0117] The present invention also provides a sepsis prediction system based on machine learning, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the steps of the aforementioned sepsis prediction system based on machine learning.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A sepsis prediction method based on machine learning, characterized in that: The method comprises the following steps: According to the ICU medical data set, a one-dimensional data set of multiple dimensions and a critical value of each dimension are obtained to classify the dimensions; the data of multiple dimensions obtained from a patient at the same acquisition time are recorded as a data sequence; multiple data sequences generated by a patient during a treatment process are recorded as data sequences during the same treatment process; in the one-dimensional data set, each data corresponds to an acquisition time; Fit all data of each dimension in the data sequence of the same treatment process to obtain the fitting slope of each data in the one-dimensional data set of each dimension; obtain the abnormal coefficient of each data in the one-dimensional data set of each dimension according to the critical value of each dimension, the fitting slope and the value of each data in the one-dimensional data set of each dimension; obtain the sepsis manifestation factor of each dimensional data according to the value and abnormal coefficient of each data in the one-dimensional data set of each dimension; According to the sepsis manifestation factor of each dimension data and the data difference between each data sequence and other data sequences in the same dimension, the distance between each data sequence and other data sequences is obtained; The clustering radius of each data sequence is obtained according to the distance between each data sequence and other data sequences during each treatment process, the sepsis manifestation factor of each dimensional data, and the abnormal coefficient of each data in each data sequence; According to the clustering radius of each data sequence and the distance between each data sequence and other data sequences, a threshold value of the number of data sequences is obtained, and all data sequences of all treatment processes are clustered into multiple clusters; Analyze different clusters, select data sequences for training sepsis prediction models, and obtain sepsis prediction models; The method of obtaining one-dimensional data sets of multiple dimensions and the critical value of each dimension based on the ICU medical data set and classifying the dimensions includes: Get the dimension data present in the Sepsis-3 standard and get One-dimensional dataset with dimensions Indicates the number of dimensions contained in the Sepsis-3 standard; The Sepsis-3 standard has different standard ranges for different dimensions. The data of the dimension is smaller than When the standard range of the dimension is at the lower limit, the patient is infected with sepsis, and the The first dimension is recorded as the first dimension, The lower limit of the standard range of the dimension is recorded as The minimum critical value of the dimension; When The data of the dimension is greater than When the upper limit of the standard range of the first dimension is reached, the patient is infected with sepsis. The dimension is recorded as the second dimension, and the The upper limit of the standard range of the dimension is recorded as The maximum critical value of the dimension; When The data of the dimension is greater than The upper limit of the standard range of the dimension or less than When the standard range of the dimension is at the lower limit, the patient is infected with sepsis, and the dimensions, recorded as the third dimension, The upper limit of the standard range of the dimension is recorded as The maximum critical value of the dimension The lower limit of the standard range of the dimension is recorded as The minimum critical value of each dimension; the maximum critical value and the minimum critical value of each dimension are recorded as the critical value of each dimension; The specific method of fitting all data of each dimension in the data sequence of the same treatment process to obtain the fitting slope of each data in the one-dimensional data set of each dimension includes: The first In a one-dimensional data set of dimensions All the treatment processes in which the data are located The data in each dimension is mapped into a two-dimensional coordinate system with the horizontal axis being the acquisition time and the vertical axis being the data value; The data points in the two-dimensional coordinate system are fitted by the least square method to obtain the In a one-dimensional data set of dimensions All the treatment processes in which the data are located The fitting curve of the data in the dimension is recorded as In the dimension A fitting curve; The first In a one-dimensional data set of dimensions The data in In the dimension The fitting slope of the corresponding data point on the fitting curve is recorded as In a one-dimensional data set of dimensions The fitted slope of the data; The specific formula for obtaining the abnormal coefficient of each data in the one-dimensional data set of each dimension according to the critical value of each dimension, the fitting slope and the value of each data in the one-dimensional data set of each dimension is as follows: When When the dimension is a first-class dimension: ； When When the dimension is a second-type dimension: ； When When the dimension is the third dimension: ； In the formula, Indicates In a one-dimensional data set of dimensions The abnormal coefficient of the data, Indicates In a one-dimensional data set of dimensions The value of the data, Indicates The minimum critical value of the dimension, Indicates In a one-dimensional data set of dimensions The fitted slope of the data, Indicates The maximum critical value of the dimension, represents the sigmoid function, represents the hyperbolic tangent function; The method of obtaining the sepsis manifestation factor of each dimension of data according to the value and abnormal coefficient of each data in the one-dimensional data set of each dimension includes the following specific methods: The first In a one-dimensional data set of dimensions The value of the data is The square of the difference between the means of all data in the one-dimensional data set of dimensions is recorded as In the dimension The difference factor of each data; The first In the dimension The difference factor of the data is In a one-dimensional data set of dimensions The product of the abnormal coefficients of the data is recorded as In the dimension The performance factor of each data; The first The normalized value of the mean of the performance factors of all data in the dimension is recorded as Sepsis manifestation factors of dimensional data; The specific method of obtaining the distance between each data sequence and other data sequences according to the sepsis expression factor of each dimensional data and the data difference between each data sequence and other data sequences in the same dimension includes: ； In the formula, Indicates The data series and The distance between the data sequences, Indicates the number of dimensions contained in the Sepsis-3 standard. Indicates In the data series The value of the dimension data, Indicates In the data series The value of the dimension data, Indicates Sepsis manifestation factors of dimensional data; The clustering radius of each data sequence is obtained according to the distance between each data sequence and other data sequences during each treatment process, the sepsis manifestation factor of each dimensional data, and the abnormal coefficient of each data in each data sequence, including the specific method of: Calculate the The data series and The difference in the acquisition time of each data series during the treatment process in which the data series are located will be The treatment process in which the data series is located is The data sequence with the largest difference in the collection time among the data sequences is recorded as The corresponding data sequence of a data sequence; No. The cluster radius of the data series The specific calculation formula is as follows: ； In the formula, No. The clustering radius of the data series, Indicates The distance between a data sequence and its corresponding data sequence, Indicates Sepsis manifestation factors of dimensional data, Indicates the number of dimensions contained in the Sepsis-3 standard. Indicates In the data series The abnormal coefficient of the dimension data; The method of obtaining the quantity threshold of the data sequence according to the clustering radius of each data sequence and the distance between each data sequence and other data sequences includes: ； In the formula, represents the quantity threshold of the data sequence, Indicates the number of data contained in each one-dimensional data set. Indicates The number of data sequences contained within the clustering radius of a data sequence, Indicates The clustering radius of the data series, Indicates The cluster radius of the data series The clustering radius of the data series, Indicates The data sequence and the The distance between the data sequences, is an exponential function with a natural constant as base, represents the normalization function; The specific method of analyzing different clusters and selecting data sequences for training a sepsis prediction model includes: The mean of the anomaly coefficients of all data in any data sequence is recorded as the local anomaly factor of the data sequence, and the mean of the local anomaly factors of all data sequences in any cluster is recorded as the average anomaly degree of the cluster; According to the average abnormality of each cluster, the clusters used to train the sepsis prediction model are screened out. The screening process is as follows: ； In the formula, Indicates the number of clusters in the clustering results. represents the number of selected clusters, Indicates the clustering result The average abnormality of the clusters, Indicates the preset selection quantity threshold; when selecting clusters, select them in descending order of average abnormality; when First greater than or equal to , the selected clusters are used as the filtered data.

2. A sepsis prediction system based on machine learning, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the sepsis prediction method based on machine learning as claimed in claim 1 are implemented.

Images