A robust risk prediction method for liver disease death, control server and medium

Abstract

The present invention discloses a robust risk prediction method for liver disease death, a control server and a medium, the method comprising: inputting liver disease data to be predicted into a death risk prediction integrated model to obtain a death risk prediction result; wherein the death risk prediction integrated model comprises a plurality of optimal risk prediction sub-models, and the diversity between the optimal risk prediction sub-models is maximized; the death risk prediction result is the optimal integrated result of the prediction results of some optimal risk prediction sub-models whose prediction results meet the set conditions. The present invention can effectively improve the accuracy of liver disease postoperative death risk prediction under abnormal data conditions.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical system-risk prediction for postoperative liver disease, and in particular to a robust risk prediction method for liver disease death, a control server and a medium. Background Art

[0002] Prediction of postoperative mortality risk in liver disease is of great significance for guiding clinical decision-making and improving patient prognosis. In the face of serious liver disease problems, although traditional treatment methods have prolonged the survival time of patients to a certain extent, the 5-year survival rate is still low. Therefore, accurate prediction of postoperative risks is crucial to improving patient survival. Through effective risk assessment, doctors can formulate more active monitoring plans and adjuvant treatments for high-risk patients, and timely detect and treat recurrence or metastasis, which is conducive to improving the overall treatment effect. Secondly, from the perspective of personalized medicine, there are significant individual differences in the condition of liver disease patients, which requires that the treatment plan can be customized according to the patient's specific risk status. Research on postoperative mortality risk prediction enables doctors to take more targeted treatment paths based on each patient's biomarkers, genetic information, clinical parameters and other data, which not only improves the success rate of treatment, but also avoids unnecessary medical interventions and reduces the economic burden and physical harm of patients. Finally, emerging technologies including deep learning and ensemble learning can process large-scale, multi-dimensional data sets and learn complex data patterns. With the increase in computing power, improvement in algorithms, increased data accessibility, and continued deepening of interdisciplinary collaboration, methods for predicting the risk of postoperative mortality in liver disease will become more accurate and personalized, bringing better clinical outcomes to patients. In summary, accurate prediction of the risk of postoperative mortality can help doctors determine which patients are more likely to benefit from more aggressive postoperative monitoring or adjuvant therapy, thereby improving patient survival. In addition, through accurate assessment of postoperative risks, more personalized treatment plans can be developed for patients and resource allocation can be optimized.

[0003] Ensemble learning is a machine learning paradigm that combines the prediction results of multiple learners to improve the accuracy, robustness and generalization ability of the overall model. The core is to optimize performance by utilizing the diversity and collective wisdom of multiple models. The advantages are as follows: 1) Ensemble learning can significantly improve prediction accuracy. By combining multiple models, it can effectively reduce overfitting and enhance the model's generalization ability for unknown data; 2) Ensemble methods utilize the diversity of different learning algorithms to make the overall model more robust and have good adaptability to various data distributions and various problems. In addition, ensemble learning can effectively integrate information. For those complex patterns and relationships that are difficult to capture with a single model, ensemble learning can better learn this information through the cooperation of multiple models; 3) Ensemble learning provides a flexible framework that can select and combine different models and algorithms according to the needs of actual problems to achieve optimal prediction performance. The above advantages make ensemble learning a powerful tool for solving complex learning tasks and is widely used in risk prediction after liver disease surgery. For example (1) Sathurthi S, Saruladha K. Prediction of liver cancer using random forest ensemble[J].

[0004] Indian Journal Of Pure&Applied Mathematics,2017,116(21):267-273; (2)Muflikhah L, Widodo N, Mahmudy W F. Prediction of liver cancer based on DNAsequence using ensemble method[C] / / 2020 3 rdInternational Seminar on Research of Information Technology and Intelligent Systems (ISRITI). IEEE, 2020: 37 - 41; (3) Krishan A, Mittal D. Ensemble liver cancer detection and classification using CT images[J]. Proceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine, 2021, 235(2): 232 - 244; (4) Zolfaghari B, Mirsadeghi L, Bibak K, et al. Cancer prognosis and diagnosis methods based on ensemble learning[J]. ACM Computing Surveys, 2023, 55(12): 1 - 34; (5) Nguyen H V, Byeon H. LIME - based ensemble machine for predicting performance status of patients with liver cancer[J]. Digital Health, 2023, 9: 1 - 17 etc.

[0005] Missing data is a common and thorny problem in studies on the prediction of mortality risk after liver surgery. First, missing data reduces the amount of data used for analysis, which in turn affects the statistical power of the study. This means that even if there are actual differences or correlations, the study may not be able to detect these effects, thereby increasing the uncertainty of the study results. Secondly, incomplete data may lead to bias in the analysis results. If the missing data is not completely random (that is, the missing data is related to the study variables), then the analysis using missing data may introduce bias, so that the research results do not accurately reflect the actual situation. Then, missing data may limit the use of specific statistical models or machine learning algorithms because some methods require a complete data set to run. This may force researchers to use other methods that may not be optimal, or require complex data interpolation techniques, which increases the complexity and uncertainty of the analysis. To alleviate the above problems, multiple imputation, maximum likelihood estimation, and model-based methods are usually used to deal with missing data, or more robust research plans are designed to reduce the impact of missing data. However, although data completion methods can alleviate the problem of missing data, they also have the following defects: 1) Introduction of human bias: The data completion process may be based on certain assumptions (such as data is randomly missing). If these assumptions do not match the actual situation, the completed data may introduce human bias, affecting the accuracy and reliability of subsequent analysis; 2) Loss of intrinsic relationships in the data: The data completion process may not fully capture the complex relationships and patterns existing in the original data, especially in liver disease data, where there may be complex biological relationships and interactions, and simply completing the data may not reflect these complexities; 3) Risk of overfitting: When building a model for data completion, there is a risk of overfitting the training data, especially when the completed model is too complex or the amount of data is insufficient. This may cause the model to be overly sensitive to the completed data, affecting the generalization ability of the model; 4) Uncertainty of results: Even if the data is completed, the accuracy and credibility of the completed data is still a problem. This uncertainty may be transmitted through the analysis, making the final prediction results uncertain; 5) Increased computational complexity: Advanced data completion methods (such as multiple imputation or model-based imputation) may require complex computational processes, increasing the time and resource consumption of the research, especially when dealing with large-scale liver disease data sets; 6) Difficult to handle missing data patterns: If the missing data are not completely random (i.e., the missing data are related to observed or unobserved values), it may be difficult to effectively handle this missing pattern using standard data completion methods, thereby affecting the accuracy of the completion and the prediction results. In summary, although data completion methods provide a possible solution to the problem of missing data in the prediction of mortality risk after liver disease surgery, when applying these methods, we need to be vigilant about potential defects and actively explore other methods to mitigate the impact of missing data, such as adopting or designing robust machine learning methods.

[0006] At present, the existing methods for predicting the risk of death after liver disease surgery have not considered how to deal with outliers in the data, resulting in the model being unable to apply to complex scenarios and unable to reflect the actual phenomenon of the risk of liver disease surgery in reality. Summary of the invention

[0007] The purpose of the present invention is to propose a robust risk prediction method for liver disease mortality, which ensures that the method can achieve high prediction accuracy in scenarios where liver disease data contain noise and outliers, thereby providing the necessary technical basis for improving the diagnostic accuracy of liver disease, promoting the development of personalized treatment strategies, and promoting in-depth research on liver disease.

[0008] The technical solution for implementing the present invention is: a robust risk prediction method for liver disease death, comprising:

[0009] Inputting the liver disease data to be predicted into the integrated model for predicting the risk of death to obtain the prediction result of the risk of death;

[0010] Wherein, the death risk prediction integrated model includes several optimal risk prediction sub-models, and the diversity among the optimal risk prediction sub-models is maximized;

[0011] The death risk prediction result is the optimal integrated result of the prediction results of some optimal risk prediction sub-models whose prediction results meet the set conditions.

[0012] Compared with the prior art, the present invention has the following significant advantages: 1) Higher noise resistance: The robust integrated model can effectively handle noise and outliers in the data, ensuring the stability and reliability of the model. Compared with traditional methods, this method is less susceptible to interference when facing data quality problems and can provide more accurate prediction results; 2) Stronger generalization ability: This method has stronger generalization ability by reducing the sensitivity to abnormal data. This means that this method is not only suitable for specific data sets, but also can maintain a high prediction accuracy on different data sets; 3) Reduced dependence on preprocessing: Due to the robustness of this method, the dependence on complex data preprocessing and cleaning steps is reduced. This simplifies the analysis process, saves research time and resources, and reduces the risk of bias introduced by human operations. In summary, the significant advantages of this method in the prediction of postoperative mortality risk of liver disease not only improve the accuracy and reliability of the research, but also provide strong support for early diagnosis, treatment selection and personalized medicine of liver disease, showing broad application potential.

[0013] The present invention is described in further detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Flowchart of a robust risk prediction method for liver disease mortality. DETAILED DESCRIPTION

[0015] Example 1

[0016] Combination Figure 1 As shown in FIG. 1 , a robust risk prediction method for liver disease mortality is described, wherein the specific steps are as follows:

[0017] Inputting the liver disease data to be predicted into the integrated model for predicting the risk of death to obtain the prediction result of the risk of death;

[0018] Wherein, the death risk prediction integrated model includes several optimal risk prediction sub-models, and the diversity among the optimal risk prediction sub-models is maximized;

[0019] The death risk prediction result is the optimal integrated result of the prediction results of some optimal risk prediction sub-models whose prediction results meet the set conditions.

[0020] In a further embodiment, the optimal risk prediction sub-model is obtained by training the risk prediction sub-model, and then a plurality of optimal risk prediction sub-models are used to generate an integrated model for predicting mortality risk. The specific training process is:

[0021] Step 1: Divide the liver disease data T (including respiratory, liver, coagulation, kidney, circulation, and nerve) into training data and test data according to an 80:20 ratio, and obtain multiple data subsets from the training data set through a self-service sampling method. The data set T is expressed as follows:

[0022]

[0023] In the formula, rs1 to rs4 are four levels of respiratory score weights, with the PaO2 / FiO2 content in the patient's exhaled air and whether the patient is mechanically ventilated as the scoring criteria. Among them, rs1 represents the weight of PaO2 / FiO2 <400mmHg and no mechanical ventilation; rs2 represents the weight of PaO2 / FiO2 <300mmHg and no mechanical ventilation; rs3 represents the weight of PaO2 / FiO2 <200mmHg and mechanical ventilation; rs4 represents the weight of PaO2 / FiO2 <100mmHg and mechanical ventilation. That is, from rs1 to rs4 corresponds to the gradual deterioration of the patient's vital signs, and accordingly, the weights are getting higher and higher, that is, rs4>rs3>rs2>rs1. The same is true for the following other weights.

[0024] ls1~ls4 are four levels of liver score weights, with bilirubin as the scoring standard, where ls1 represents the weight of bilirubin belonging to [20, 32] (i.e. 20umol / L≤bilirubin≤32umol / L); ls2 represents the weight of bilirubin belonging to [33, 101]; ls3 represents the weight of bilirubin belonging to [102, 204]; ls4 represents the weight of bilirubin>204umol / L.

[0025] cos1 to cos4 are four levels of coagulation score weights, with platelet count as the scoring standard, where cos1 means that the platelet count belongs to [101, 105] (i.e. 101×10 9 / L≤platelet count≤105×10 9 / L); cos2 represents the weight of platelet count belonging to [51, 100]; cos3 represents the weight of platelet count belonging to [21, 50]; cos4 represents the weight of platelet count <21×10 9 / L weight.

[0026] Ks1~ks4 are the four levels of kidney score weights, with creatinine and 24-hour urine volume as scoring indicators, among which, Ks1 represents the weight of creatinine belonging to [110, 170] (i.e. 110umol / L≤creatinine≤170umol / L); Ks2 represents the weight of creatinine belonging to [171, 299]; Ks3 represents the weight of creatinine belonging to [300, 440] and 24-hour urine volume belonging to [20, 500] (i.e. 20ml≤24-hour urine volume≤500ml); Ks4 represents the weight of creatinine>440umol / L and 24-hour urine volume<200ml.

[0027] Cys1 to Cys4 are four levels of cyclic scoring weights, with mean arterial pressure, dopamine dose, epinephrine dose, norepinephrine dose, and dobutamine as scoring criteria, among which Cys1 represents the weight of mean arterial pressure <70 mmHg; Cys2 represents the weight of dopamine dose ≤5 ug / kg / min or the use of dobutamine; Cys3 represents the weight of dopamine dose >5 ug / kg / min or epinephrine dose ≤0.1 ug / kg / min or norepinephrine dose ≤0.1 ug / kg / min; Cys4 represents the weight of dopamine dose >15 ug / kg / min or epinephrine dose >0.1 ug / kg / min or norepinephrine dose >0.1 ug / kg / min.

[0028] ns1 to ns4 are four levels of neurological score weights, with the Glasgow Coma Scale (GCS) score as the scoring indicator, where ns1 represents the weight of the GCS score belonging to [13, 14]; ns2 represents the weight of the GCS score belonging to [10, 12]; ns3 represents the weight of the GCS score belonging to [6, 9]; and ns4 represents the weight of the GCS score <6.

[0029] Step 2: Build a robust risk prediction sub-model, and then input the training data subset into the sub-model for model training. Each training data subset trains a risk prediction sub-model, such as Figure 1 As shown, specifically:

[0030]

[0031] where p∈(0,2], ||·|| p For L p Norm distance; c m and q m are penalty coefficients and slack variables respectively; matrix A m and B m are the m-th sample (representing the postoperative survival data of liver disease) and the remaining samples except the m-th sample (representing the postoperative death data of liver disease); w m and b m is the weight vector and bias of the m-th class sample; e 1m and e 2m is a unit column vector.

[0032] The slack variable q m =e 2m +(B m w m +e 2m b m ) into formula (1), we can get:

[0033]

[0034] Rewrite formula (2) as follows:

[0035]

[0036] According to the mathematical symbol H m =(A m e 1m ), G m =(B m e 2m ) and z m =(w m , b m ). Formula (3) is transformed into the following form:

[0037]

[0038] Formula (4) involves the absolute value optimization problem of variables, which makes it difficult to solve directly, so the trace function sign(·) is introduced. Details are as follows:

[0039] |H m z m |=diag(sign(H m z m ))(H m z m ) (5)

[0040] |e 2m +G m z m |=diag(sign(e 2m +G m z m ))(e 2m +G m z m ) (6)

[0041] Solve formula (4) for z m The partial derivatives of are as follows:

[0042]

[0043] Rewrite formula (7) as follows:

[0044]

[0045] Since the diagonal matrix and The i-th and j-th diagonal elements of and Formula (8) is optimized into the following form:

[0046]

[0047] The augmented vector z can be obtained by formula (9): m ,as follows:

[0048]

[0049] The regularization term εI is introduced into formula (10) to solve the matrix singularity problem. By obtaining the weight vector w m and deviation b m The optimal classification plane is as follows:

[0050] x T w m +b m=0. (11)

[0051] The decision function (12) is used to calculate the perpendicular distance between the data point and the two hyperplanes. Then, the data point is classified into the mth class according to the calculated distance.

[0052]

[0053] Step 3: Update the parameters of each sub-model through the self-developed iterative algorithm, and obtain the optimal weight w when the convergence conditions are met. m , generate the optimal sub-model. The specific steps are:

[0054] Step 3.1: Initialize the number of iterations t = 1, the step size p = 0.5, and randomly assign the augmented vector Then according to Compute the diagonal matrix and The i-th diagonal element is

[0055] The jth diagonal element is Where m1 is the number of positive samples, m2 is the number of negative samples, and h i is the matrix H m =(A m e 1m ), g j is the matrix G m =(B m e 2m ), e 2mj is the column vector e 2m The jth element of . The positive samples represent the survival data after liver disease surgery; the negative samples represent the death data after liver disease surgery.

[0056] Step 3.2: According to the diagonal matrix and Calculate the augmented vector The specific calculation formula is:

[0057]

[0058] Among them, H m =(A m e 1m ), G m =(B m e 2m ), ε and I are the regularization coefficient and the unit matrix respectively.

[0059] Step 3.3: According to Update the diagonal matrix and The i-th diagonal element of The j-th diagonal element of is specifically:

[0060]

[0061] Step 3.4: Calculate the objective function value of the sub-model like And t≤50, then update the objective function value t=t+1,p=p+0.5,return to step 3.2,otherwise generate the optimal sub-model and output the augmented vector

[0062] Step 4: Use the alternating direction multiplier method to solve the designed diversity metric function and generate a robust integration model that maximizes the diversity among components. The specific steps are:

[0063] Step 4.1: Construct the diversity metric function F(w), specifically: Among them, p k and p l The kth row and lth column of the diagonal matrix D are the prediction results of the kth and lth (k, l = 1, 2, ..., L) classifiers respectively. kl =2(p k -p l ) T (p k -p l ).

[0064] Step 4.2: Solve the diversity metric function F(w) by the alternating direction multiplier method. When w m and b m When fixed, F(w) is transformed into the following form:

[0065]

[0066] Where w = [w1, w2, ..., w L ] T is the weight coefficient. The Lagrangian function of formula (13) is constructed as follows:

[0067] L2(w)=λw T Dw-α(e T w-1), (14)

[0068] Among them, α is the Lagrange multiplier, and e is a column vector that is all 1. By solving equation (14), the weight coefficient w can be obtained as follows:

[0069]

[0070] Among them, the Lagrange multiplier α = -(2λ / e T D -1 e).

[0071] Step 4.3: Generate a robust integration model that maximizes the diversity among components, specifically:

[0072]

[0073] Among them, λ is the equilibrium parameter of the model.

[0074] When the model is optimized Achieve a balance with F(w), that is, the diversity among the components in the integrated model is maximized, which is conducive to improving the prediction accuracy of the model.

[0075] The method for determining the mortality risk prediction result is as follows: input the liver disease data to be predicted into the generated robust integrated model, and then obtain the corresponding prediction results, and remove the poor prediction results through the developed trimming algorithm to give the final prediction value. The specific steps are:

[0076] Step 5.1, initialize the clipping factor γ, set l = 1, ..., L, L is the number of optimal risk prediction sub-models;

[0077] Step 5.2: Calculate the median f of the prediction results of the optimal risk prediction sub-model i , and calculate the average value f of the prediction results of all optimal risk prediction sub-models avg ;

[0078] Step 5.3, if (f l / f avg )>γ, then delete the prediction results of the sub-model, l=l+1, and return to step 5.2; otherwise, integrate the prediction results of the optimal risk prediction sub-model, and then output the integrated optimal prediction results.

[0079] The present invention has the following advantages:

[0080] 1) Improve prediction accuracy: Liver disease data inevitably contain noise and outliers, such as measurement errors, deviations in sample processing, etc. This solution can effectively deal with these problems, reduce the impact of noise and outliers, and thus improve the accuracy and reliability of the prediction model. In addition, this solution reduces the reliance on complex data preprocessing to a certain extent, simplifies the research process, and accelerates research progress; 2) Enhance the generalization ability of the model: This solution reduces the reliance on extreme values ​​and distribution assumptions, so that the model can better reflect the true distribution of the data, enhance the model's prediction ability for unknown data, and improve the generalization ability of the model; 3) Promote personalized treatment: There are significant biological differences between patients with liver disease. This solution can more accurately distinguish different subtypes or prognosis groups, providing support for the realization of precision medicine and personalized treatment strategies; 4) Support clinical decision-making: Accurate liver disease prediction is crucial to determine the best treatment plan, evaluate prognosis, and guide clinical decision-making. This solution can provide more reliable prediction results to help doctors make more reasonable treatment choices; 5) Promote in-depth research on liver disease: This solution can reveal the potential mechanisms and influencing factors of the occurrence and development of liver disease, and provide new perspectives and methods for studying the molecular biological basis and pathogenesis of liver disease. In summary, in the prediction of postoperative mortality risk for liver disease, this scheme can not only improve the quality of research and the credibility of results, but also provide important scientific basis for early diagnosis, treatment selection and prognosis evaluation of liver disease, which is of great significance for improving the survival rate and quality of life of patients with liver disease.

[0081] Example 2

[0082] A control server is presented in the form of a general-purpose computing device, including but not limited to: one or more processors or processing units, a system memory, and a bus connecting the system memory and the processing unit.

[0083] Bus refers to one or more of several types of bus structures, including a memory bus or a memory controller peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnect (PCI) bus.

[0084] The control server includes a variety of computer system readable media. These media can be any available media that can be accessed by the control server, including volatile and non-volatile media, removable and non-removable media.

[0085] The system memory may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The control server may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system may be used to read and write non-removable, non-volatile magnetic media (commonly referred to as "hard drives"). A disk drive for reading and writing removable non-volatile disks (such as "floppy disks"), and an optical disk drive for reading and writing removable non-volatile optical disks (such as CD-ROMs, DVD-ROMs or other optical media) may be provided. In these cases, each drive may be connected to the bus via one or more data medium interfaces. The system memory may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of various embodiments of the present invention.

[0086] A program / utility having a set (at least one) of program modules may be stored, for example, in a system memory, such program modules including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules generally perform the functions and / or methods of the embodiments described herein.

[0087] The control server may also communicate with one or more external devices, such as keyboards, pointing devices, displays, etc., and may also communicate with one or more devices that enable users to interact with the device, and / or communicate with any device (such as a network card, a modem, etc.) that enables the control server to communicate with one or more other computing devices. Such communication may be performed through an input / output (I / O) interface. In addition, the server may also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter. The network adapter communicates with other modules of the control server through a bus. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the control server, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems. The processing unit executes various functional applications and data processing by running programs stored in the system memory, such as implementing the robust risk prediction method for liver disease death provided in an embodiment of the present invention.

[0088] Example 3

[0089] A computer-readable storage medium stores a computer program, which, when executed by a processor, can implement any of the robust liver disease death risk prediction methods in the above-mentioned embodiments.

[0090] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only storage (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0091] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0092] Program code embodied on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to:

[0093] Wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0094] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

Claims

1. A robust risk prediction method for liver disease mortality, characterized in that: include: Inputting the liver disease data to be predicted into the integrated model for predicting the risk of death to obtain the prediction result of the risk of death; The liver disease data T are specifically: Among them, rs1~rs4 are four levels of respiratory score weights, with the PaO2 / FiO2 content index in the exhaled air and whether mechanical ventilation is used as the scoring standard; ls1~ls4 are four levels of liver score weights, with the bilirubin index as the scoring standard; cos1~cos4 are four levels of coagulation score weights, with the platelet count index as the scoring standard; ks1~ks4 are four levels of kidney score weights, with creatinine and 24-hour urine volume as the scoring indicators; cys1~cys4 are four levels of circulation score weights, with mean arterial pressure, dopamine dose, epinephrine dose, norepinephrine dose, and dobutamine as the scoring standards; ns1~ns4 are four levels of neurological score weights, with the Glasgow Coma Scale as the scoring indicator; The death risk prediction integrated model includes several optimal risk prediction sub-models, and the diversity among the optimal risk prediction sub-models is maximized; The risk prediction sub-model is trained using the liver disease data training set to obtain the optimal risk prediction sub-model, wherein the risk prediction sub-model is specifically: where p∈(0,2], ||·|| p For L p Norm distance, the p in the upper right corner represents the pth power; c m and q m are penalty coefficients and slack variables respectively; matrix A m and B m is the mth class sample and the remaining samples except class m; w m and b m is the weight vector and bias of the mth class sample; e 1m and e 2m is a unit column vector, the mth sample is the survival data after liver disease surgery, and the remaining samples except the mth sample are the death data after liver disease surgery; The specific method to obtain the optimal risk prediction sub-model is: Step 3.1: Initialize the number of iterations t = 1, the step size p = 0.5, and randomly assign the augmented vector according to Compute the diagonal matrix and The i-th diagonal element is The jth diagonal element is Where m1 is the number of positive samples, m2 is the number of negative samples, and h i is the matrix H m =(A m e 1m ), g j is the matrix G m =(B m e 2m ), e 2mj is the column vector e 2m The jth element of m and b m is the weight vector and bias of the mth class sample; Step 3.2: According to the diagonal matrix and Calculate the augmented vector The specific calculation formula is: Among them, ε and I are the regularization coefficient and the unit matrix, respectively, c m is the penalty coefficient; Step 3.3: According to Update the diagonal matrix and The i-th diagonal element of The j-th diagonal element of is specifically: Step 3.4: Calculate the objective function value of the risk prediction sub-model like And t≤50, then update the objective function value t=t+1,p=p+0.5,return to step 3.2,otherwise generate the optimal risk prediction sub-model and output the augmented vector The way to maximize the diversity among the optimal risk prediction sub-models is as follows: Step 4.1: Construct the diversity metric function F(w), specifically: Among them, p k and p l are the prediction results of the kth and lth optimal risk prediction sub-models, respectively, k, l = 1, 2, ..., L, L is the number of optimal risk prediction sub-models, w = [w1, w2, ..., w L ] T is the weight coefficient, and the kth row and lth column of the diagonal matrix D is d kl =2(p k -p l ) T (p k -p l ); Step 4.2: Solve the diversity metric function F(w) by the alternating direction multiplier method, specifically: w = -(α / λ)D -1 e, α are Lagrange multipliers, and e is a column vector that is all 1; Step 4.3: Generate a robust ensemble model that maximizes the diversity among the optimal risk prediction sub-models, i.e., a mortality risk prediction ensemble model, specifically: Among them, λ is the equilibrium parameter of the model, w k is the weight coefficient for each sub-model; When the robust ensemble model is optimized, Achieve balance with F(w), that is, the diversity among the optimal risk prediction sub-models in the integrated model is maximized; The death risk prediction result is the optimal integrated result of the prediction results of some optimal risk prediction sub-models whose prediction results meet the set conditions.

2. The robust risk prediction method for liver disease death according to claim 1, characterized in that: The risk prediction sub-model is a classifier.

3. The robust risk prediction method for liver disease death according to claim 1, characterized in that: The method for determining the mortality risk prediction results is as follows: Step 5.1, initialize the clipping factor γ, set l = 1, ..., L, L is the number of optimal risk prediction sub-models; Step 5.2: Calculate the median f of the prediction results of the optimal risk prediction sub-model l , and calculate the average value f of the prediction results of all optimal risk prediction sub-models avg ; Step 5.3, if (f l / f avg )>γ, then delete the prediction results of the sub-model, l=l+1, and return to step 5.2; otherwise, integrate the prediction results of the optimal risk prediction sub-model, and then output the integrated optimal prediction results.

4. The robust risk prediction method for liver disease death according to claim 1, characterized in that: The optimal integrated result of the prediction results of some optimal risk prediction sub-models refers to the mean result of the prediction results of some optimal risk prediction sub-models.

5. A control server, characterized in that: include: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the robust risk prediction method for death from liver disease as described in any one of claims 1-4.

6. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the robust risk prediction method for liver disease death as described in any one of claims 1-4 is implemented.

Images