Prediction method and system of super-progress prediction model based on machine learning

Through the hyper-progress prediction model based on machine learning, the problem that the existing logistic regression model is not objective enough and has low accuracy in clinical practice, and more efficient and accurate clinical data processing and prediction results are achieved.

CN119066415APending Publication Date: 2024-12-03ZHONGDA HOSPITAL SOUTHEAST UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410947593.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

In clinical practice, the existing logistic regression model is not objective enough, has low accuracy, and cannot be effectively applied to clinical work.

Method used

Using a machine learning-based hyperprogress prediction model, obtain clinical data for preprocessing and missing value processing, use Lasso regression model for feature screening, and input data into multiple machine learning models (such as logistic regression, random forest, extreme gradient enhancement tree, etc.) for training and performance evaluation, and filter out models with the best prediction performance for prediction.

Benefits of technology

The speed and efficiency of model construction are improved, the authenticity and accuracy of the prediction results are ensured, the contribution distribution of each feature can be better reflected, and the contribution of each feature in a certain sample is presented in a targeted manner.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066415A_ABST
    Figure CN119066415A_ABST
Patent Text Reader

Abstract

The invention discloses a prediction method and system for a super-progress prediction model based on machine learning, and relates to the technical field of machine learning, and the method comprises the steps: obtaining clinical related data, carrying out the preprocessing of the clinical related data, obtaining the processed clinical related data, and dividing the processed clinical related data into a training set and a test set; inputting the training set into a pre-established Lasso regression model to screen predictive factors, outputting to obtain a screened training set, and respectively inputting the screened training set into a plurality of pre-established machine learning models to train to obtain a plurality of trained machine learning models; and performing performance evaluation on the multiple trained machine learning models according to set indexes, screening out the machine learning model with the optimal prediction efficiency, inputting the test set into the machine learning model with the optimal prediction efficiency, and outputting to obtain a prediction result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a prediction method and system for a hyperprogression prediction model based on machine learning. Background Art

[0002] The logistic regression used in the existing technology is based on the assumption of linear correlation and is a parametric model that calculates scores based on fixed weights. However, the actual clinical situation is more complicated, which will lead to the model construction process being not objective enough, and the model's accuracy is not high, and it cannot be effectively applied to clinical work. Summary of the invention

[0003] In order to solve the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide a prediction method and system for a hyperprogression prediction model based on machine learning.

[0004] In a first aspect, the purpose of the present invention can be achieved by the following technical solution: a prediction method based on a hyperprogression prediction model of machine learning, the method comprising the following steps:

[0005] Acquire clinically relevant data, pre-process the clinically relevant data to obtain processed clinically relevant data, and divide the processed clinically relevant data into a training set and a test set;

[0006] The training set is input into a pre-established Lasso regression model to screen predictive factors, and the screened training set is outputted. The screened training set is respectively input into a plurality of pre-established machine learning models for training, and a plurality of trained machine learning models are obtained;

[0007] According to the set indicators, the performance of multiple trained machine learning models is evaluated, and the machine learning model with the best prediction performance is screened out. The test set is input into the machine learning model with the best prediction performance, and the prediction result is output.

[0008] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: preprocessing the clinically relevant data to process missing data values, the process is as follows:

[0009] Setting a first ratio threshold and a second ratio threshold; wherein the first ratio threshold is greater than the second ratio threshold;

[0010] Calculate the missing ratio of clinically relevant data. If the missing ratio exceeds the first ratio threshold, delete the corresponding clinically relevant data without further processing.

[0011] If the missing proportion exceeds the second proportion threshold but does not exceed the first proportion threshold, the missing clinically relevant data are uniformly marked as Unknown, and the non-missing clinically relevant data are classified according to the reference value range and converted into categorical variables in combination with the missing value group;

[0012] If the missing proportion does not exceed the second proportion threshold, the clinically relevant data are processed using the missing value interpolation method.

[0013] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the pre-established Lasso regression model adds an L1 regularization term on the basis of the linear regression model to make many parameters become zero to perform a feature screening function.

[0014] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: the multiple pre-established machine learning models include a logistic regression model LR, a random forest model RF, an extreme gradient boosting tree model XGBoost, a category feature gradient enhancement model CatBoost, and a support vector machine model SVM.

[0015] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: the set indicators for evaluating the performance of multiple trained machine learning models include area under the curve AUC, accuracy, sensitivity, specificity, positive predictive value PPV and negative predictive value NPV.

[0016] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the accuracy calculation formula is as follows:

[0017]

[0018] The sensitivity calculation formula is as follows:

[0019]

[0020] The specificity calculation formula is as follows:

[0021]

[0022] The positive predictive value (PPV) is calculated as follows:

[0023]

[0024] The negative predictive value (NPV) is calculated as follows:

[0025]

[0026] The area under the curve AUC is the area under the ROC curve and the coordinate axis;

[0027] Wherein, accuracy is accuracy, sensitivity is sensitivity, specificity is specificity, TP is true positive result value, TN is true negative result value, FP is false positive result value, and FN is false negative result value.

[0028] In a second aspect, in order to achieve the above-mentioned object, the present invention discloses a prediction system of a hyperprogression prediction model based on machine learning, comprising:

[0029] A data processing module is used to obtain clinically relevant data, pre-process the clinically relevant data, obtain processed clinically relevant data, and divide the processed clinically relevant data into a training set and a test set;

[0030] A model training module is used to input the training set into a pre-established Lasso regression model to screen predictive factors, output a screened training set, and input the screened training set into a plurality of pre-established machine learning models for training to obtain a plurality of trained machine learning models;

[0031] The model prediction module is used to evaluate the performance of multiple trained machine learning models according to set indicators, screen out the machine learning model with the best prediction performance, input the test set into the machine learning model with the best prediction performance, and output the prediction results.

[0032] In conjunction with the second aspect, in certain implementations of the second aspect, the system further includes: preprocessing the clinically relevant data in the data processing module to perform data missing value processing, the process is as follows:

[0033] Setting a first ratio threshold and a second ratio threshold; wherein the first ratio threshold is greater than the second ratio threshold;

[0034] Calculate the missing ratio of clinically relevant data. If the missing ratio exceeds the first ratio threshold, delete the corresponding clinically relevant data without further processing.

[0035] If the missing proportion exceeds the second proportion threshold but does not exceed the first proportion threshold, the missing clinically relevant data are uniformly marked as Unknown, and the non-missing clinically relevant data are classified according to the reference value range and converted into categorical variables in combination with the missing value group;

[0036] If the missing proportion does not exceed the second proportion threshold, the clinically relevant data are processed by the missing value interpolation method;

[0037] The Lasso regression model pre-established in the model training module adds an L1 regularization term to the linear regression model, making many parameters zero to perform feature screening;

[0038] The model training module includes a variety of pre-built machine learning models, including logistic regression model LR, random forest model RF, extreme gradient boosting tree model XGBoost, category feature gradient boosting model CatBoost, and support vector machine model SVM;

[0039] The indicators set in the model prediction module to evaluate the performance of various trained machine learning models include area under the curve AUC, accuracy, sensitivity, specificity, positive predictive value PPV and negative predictive value NPV;

[0040] The accuracy calculation formula in the model prediction module is as follows:

[0041]

[0042] The sensitivity calculation formula is as follows:

[0043]

[0044] The specificity calculation formula is as follows:

[0045]

[0046] The positive predictive value (PPV) is calculated as follows:

[0047]

[0048] The negative predictive value (NPV) is calculated as follows:

[0049]

[0050] The area under the curve AUC is the area under the ROC curve and the coordinate axis;

[0051] Wherein, accuracy is accuracy, sensitivity is sensitivity, specificity is specificity, TP is true positive result value, TN is true negative result value, FP is false positive result value, and FN is false negative result value.

[0052] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, the prediction method of the hyper-progression prediction model based on machine learning as described above is adopted.

[0053] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a computer-readable storage medium is disclosed, in which a computer program is stored. When the computer program is loaded and executed by a processor, the prediction method of the hyperprogression prediction model based on machine learning as described above is adopted.

[0054] Beneficial effects of the present invention:

[0055] In the process of model construction, machine learning can efficiently process multidimensional data, and there are no strict requirements for data preprocessing, which can improve the speed and efficiency of model construction. At the same time, in terms of model effectiveness, machine learning is more flexible in processing data, which can better ensure the authenticity and accuracy of the prediction results, and can not only reflect the contribution distribution of each feature from the overall perspective of the model, but also specifically present the contribution of each feature in a certain sample. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0057] Figure 1 It is a schematic flow chart of the method of the present invention;

[0058] Figure 2 It is a schematic diagram of the change curve of the Lasso regression coefficient with Log(λ) of the present invention;

[0059] Figure 3 Schematic diagram of AUC curves of the training set and test set of the CAT model of the present invention;

[0060] Figure 4 It is a SHAP interpretable analysis honeycomb diagram of the CAT model of the present invention;

[0061] Figure 5 It is a schematic diagram of the SHAP interpretable analysis waterfall of the CAT model A of the present invention;

[0062] Figure 6 It is a schematic diagram of the SHAP interpretable analysis waterfall of the CAT model of B of the present invention;

[0063] Figure 7 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] Embodiment 1:

[0066] like Figure 1 As shown, the prediction method of the super-progression prediction model based on machine learning includes the following steps:

[0067] S101: Acquire clinically relevant data, pre-process the clinically relevant data to obtain processed clinically relevant data, and divide the processed clinically relevant data into a training set and a test set;

[0068] The clinical data is preprocessed to handle missing values. The process is as follows:

[0069] Setting a first ratio threshold and a second ratio threshold; wherein the first ratio threshold is greater than the second ratio threshold;

[0070] Calculate the missing ratio of clinically relevant data. If the missing ratio exceeds the first ratio threshold, delete the corresponding clinically relevant data without further processing.

[0071] If the missing proportion exceeds the second proportion threshold but does not exceed the first proportion threshold, the missing clinically relevant data are uniformly marked as Unknown, and the non-missing clinically relevant data are classified according to the reference value range and converted into categorical variables in combination with the missing value group;

[0072] If the missing ratio does not exceed the second ratio threshold, the clinically relevant data are processed by the missing value interpolation method. Among the existing multiple imputation methods, multivariate imputation by chained equations (MICE) based on the random forest model is a method that can flexibly handle continuous variables and multi-type categorical variables. Therefore, in this study, the "mice package" in the R language was chosen to use the random forest-based multiple imputation algorithm to process the data. In addition, after the imputation is completed, a sensitivity analysis of the data before and after the imputation is required.

[0073] In this embodiment, the first ratio threshold is set to 60%, and the second ratio threshold is set to 10%;

[0074] When describing the missing data statistically, there was no clinical data with a missing ratio exceeding 60%;

[0075] The missing ratio did not exceed 60%, but there were 7 clinical data items with a missing ratio of more than 10%, namely free triiodothyronine (31, 23.13%), free thyroxine (32, 23.88%), thyroid stimulating hormone (34, 25.37%), triiodothyronine (58, 43.28%), thyroxine (58, 43.28%), calcitonin (59, 44.03%), and thyroglobulin (59, 44.03%);

[0076] There are 16 clinical data whose missing ratio does not exceed 10%, namely diabetes (1, 0.75%), indirect bilirubin (2, 1.49%), glucose (1, 0.75%), triglycerides (8, 5.97%), total cholesterol (8, 5.97%), high-density lipoprotein cholesterol (10, 7.46%), low-density lipoprotein cholesterol (10, 7.46%), apolipoprotein A-1 (10, 7.46%), apolipoprotein B (10, 7.46%), lipoprotein a (10, 7.46%), prothrombin activity (2, 1.49%), antithrombin activity (1, 0.75%), fibrin degradation products (2, 1.49%), alpha-fetoprotein (2, 1.49%), carcinoembryonic antigen (7, 5.22%), and carbohydrate chain antigen 19-9 (7, 5.22%). After using the MICE multiple imputation method to fill missing values, sensitivity analysis was performed on the data before and after interpolation, and the results are shown in Table 1. The results showed that the P values ​​before and after interpolation were greater than 0.05, indicating that there was no significant difference in the data before and after interpolation, and subsequent analysis could be performed.

[0077] Table 1 Missing data statistics and sensitivity analysis

[0078]

[0079]

[0080] Note: Missing variables are presented as “missing number (proportion)”; categorical variables are presented as χ 2 The results were presented as “n(%)” and the independent sample t-test was used for continuous variables, and the results were presented as “mean ± standard deviation”.

[0081] The processed clinically relevant data were divided into training set and test set with a ratio of 6 to 4;

[0082] Specifically, the dataset was divided into a training set (n=80) and a test set (n=54) in a ratio of 6:4 using completely random sampling. The differences between the two datasets were analyzed using data on gender, age, BMI, past medical history, PS score, Child-Pugh grade, CNLC stage of HCC, PVTT status, previous liver resection, and other combined treatments. The results (Table 2) showed that there were no significant differences in the above baseline characteristics between the training set and the test set (P value > 0.05), indicating that there was no statistical difference in the baseline conditions of the patients between the two datasets, and subsequent model establishment and verification could be performed.

[0083] Table 2 Baseline characteristics of the training and test sets

[0084]

[0085]

[0086] Note: 2 : Chi-square test, expressed as “n(%)”; -: Fisher’s exact test, expressed as “n(%)”; Z: Mann-Whitney rank-sum test, expressed as “median (quartile)”.

[0087] S102: inputting the training set into a pre-established Lasso regression model to screen predictive factors, outputting a screened training set, and inputting the screened training set into a plurality of pre-established machine learning models for training, to obtain a plurality of trained machine learning models;

[0088] Perform Lasso regression in the training set and draw a regression coefficient path diagram ( Figure 2 ). The optimization goal of Lasso regression is to find the best coefficient by minimizing the loss function, and the loss function consists of two parts, one is the error between the predicted value and the actual value, and the other is the L1 regularization term. This L1 regularization term is the regularization parameter Lambda value (λ) multiplied by the sum of the absolute values ​​of all coefficients. Figure 2 It is the curve of the Lasso regression coefficient changing with Log(λ). Each line shows the changing trajectory of each independent variable coefficient. The lower horizontal axis is Log(λ), and the upper horizontal axis is the number of independent variables with non-zero coefficients in the model corresponding to Log(λ). Figure 2 As can be seen, as the parameter log Lambda increases, the regression coefficient continues to converge and eventually converges to 0. The variables that have not shrunk to 0 are more important.

[0089] The larger the Lambda value, the stronger the penalty for the coefficient of the model, which will lead to more parameter estimates of zero, that is, more features are ignored; conversely, more variables will be retained, resulting in overfitting of the model. Therefore, the choice of Lambda value is crucial to the results of Lasso regression. Therefore, in this study, the largest Lambda value with a mean error within one standard deviation was selected, and the results were printed in R, where the coefficient represents the coefficient of each feature under this Lambda value, and 0 means it is not important, and the features that are not 0 are screened. Finally, we used Lasso regression analysis to screen out 95 features, namely age (age), LYM (lymphocyte count), ALB (albumin), AST (aspartate aminotransferase), BUN (urea), LPa (lipoprotein a), TT (thrombin time), CA19-9 (glycoprotein antigen 19-9), (FDP) fibrin degradation products, a total of 9 features, for the construction of subsequent clinical prediction models.

[0090] Model formula:

[0091]

[0092] Age, LYM, CA19-9, AST, BUN, TT, LPa, FDP, ALB

[0093] A variety of pre-built machine learning models include logistic regression model LR, random forest model RF, extreme gradient boosting tree model XGBoost, category feature gradient boosting model CatBoost, and support vector machine model SVM.

[0094] S103: Evaluate the performance of multiple trained machine learning models according to set indicators, select the machine learning model with the best prediction performance, input the test set into the machine learning model with the best prediction performance, and output the prediction result.

[0095] The model performance is evaluated in the training set and the test set respectively. The results are shown in Table 3.

[0096] Table 2.6 Evaluation of 5 machine learning algorithms in training and test sets

[0097]

[0098]

[0099] Note: LR: Logistic regression model, RF: Random forest model, XGBoost: Extreme gradient boosting tree, CatBoost: Category feature gradient boosting model, SVM: Support vector machine

[0100] Table 3 shows that among the five machine learning models in the test set, CAT has the highest test set AUC value (0.777), accuracy (0.870), specificity (0.750), PPV (0.953) and NPV (0.545) among the five models, and its sensitivity is relatively good (0.891), which has the best effect. Finally, we choose CAT as the final model, and its AUC curve is as follows: Figure 3 shown.

[0101] The accuracy calculation formula is as follows:

[0102]

[0103] The sensitivity calculation formula is as follows:

[0104]

[0105] The specificity calculation formula is as follows:

[0106]

[0107] The positive predictive value (PPV) is calculated as follows:

[0108]

[0109] The negative predictive value (NPV) is calculated as follows:

[0110]

[0111] The area under the curve AUC is the area under the ROC curve and the coordinate axis;

[0112] Wherein, accuracy is accuracy, sensitivity is sensitivity, specificity is specificity, TP is true positive result value, TN is true negative result value, FP is false positive result value, and FN is false negative result value.

[0113] The prediction result is whether HPD occurs.

[0114] The evaluation criteria for HPD are mainly based on the following four indicators: RECIST 1.1, TGR, TGK and TTF.

[0115] Among them, the first three indicators are calculated by the sum of the longest diameters (SLD) of the target lesions. To obtain the SLD value, the target lesions of each hepatocellular carcinoma patient must be determined according to the RECIST1.1 standard (select 5 measurable lesions, and each organ can select up to 2 as target lesions). Add the long diameters of the target lesions to get a long diameter sum, and then add it to the short diameter of the lymph nodes exceeding 15 mm to get the SLD value.

[0116] The first indicator was the percentage of SLD increase compared with the start of treatment, ie, ΔSLD / SLD0×100%.

[0117] TGR represents the variation index of tumor volume per unit time under the assumption that tumor growth follows an exponential law. The calculation formula is as follows:

[0118]

[0119] TGK represents the variation index of SLD per unit time, and the calculation formula is as follows:

[0120]

[0121] In this study, the SLD values ​​were uniformly measured by the same radiologist with many years of experience based on imaging images to ensure uniformity of observation standards.

[0122] HPD in HCC patients was defined as meeting three or more of the following criteria between baseline and first radiographic evaluation:

[0123] (1) SLD increased by more than 50%;

[0124] (2) TGR or TGK increased more than 2 times, that is, TGR t / TGR0>2 or TGK t / TGK0>2;

[0125] (3) TTF < 2 months;

[0126] (4) ECOG score decreased by >2 points within 2 months after treatment;

[0127] The evaluation criteria for HPD are mainly based on the following four indicators: RECIST 1.1, TGR, TGK and TTF.

[0128] Among them, the first three indicators are calculated by the sum of the longest diameters (SLD) of the target lesions. To obtain the SLD value, the target lesions of each hepatocellular carcinoma patient must be determined according to the RECIST1.1 standard (select 5 measurable lesions, and each organ can select up to 2 as target lesions). Add the long diameters of the target lesions to get a long diameter sum, and then add it to the short diameter of the lymph nodes exceeding 15 mm to get the SLD value.

[0129] The first indicator was the percentage of SLD increase compared with the start of treatment, ie, ΔSLD / SLD0×100%.

[0130] TGR represents the variation index of tumor volume per unit time under the assumption that tumor growth follows an exponential law. The calculation formula is as follows:

[0131]

[0132] TGK represents the variation index of SLD per unit time, and the calculation formula is as follows:

[0133]

[0134] SHAP (SHapleyAdditive exPlanations) is selected for interpretable analysis, and the model is visualized as a whole and in part through honeycomb charts and waterfall charts. SHAP is a model interpretation method based on game theory. It introduces the concept of Shapley value and reflects the contribution of each feature to the final prediction result by calculating the contribution of each feature to the prediction. For each prediction sample, the model generates a prediction value, and the SHAP value is the value assigned to each feature in the sample.

[0135] 3.6.1 CAT model overall visualization

[0136] Honeycomb diagram Figure 4 ) is a method for global interpretation, that is, the expected behavior of the overall distribution of input variable values ​​contributing to the model, which can reveal not only the relative importance of features, but also their actual relationship with the predicted results. In the honeycomb diagram, the horizontal axis is the SHAP value, which represents the distribution of the contribution of features to the model, and positive and negative values ​​represent positive and negative impacts on the model; the vertical axis represents the features, and the vertical axis sorts the features according to the average absolute value of the SHAP value of all samples, that is, from top to bottom, the importance of these features to the model is from large to small; each point represents a sample, and the overlapping points are stacked vertically to show the distribution of the SHAP value of each feature; yellow and purple correspond to high and low values, respectively.

[0137] exist Figure 4It can be seen that in the CAT model, FDP (fibrin degradation product) contributes the most, followed by CA19-9 (glycoprotein antigen 19-9), LYM (lymphocyte count), AST (aspartate aminotransferase), and BUN (urea) are also important to the model, while age (age), TT (thrombin time), and LPa (lipoprotein a) are relatively less important to the model.

[0138] Specifically, the present invention is further described below by way of embodiments:

[0139] CAT model single sample visualization

[0140] Waterfall Chart( Figure 5 , 6 ) is a method used to explain a single sample, that is, to describe the prediction result of a sample based on this model and reflect the contribution of each feature to this prediction. We selected two patients in the test set (patient A and patient B) to use the CATBoost model for prediction and used the SHAP waterfall chart to perform interpretable analysis of the single sample prediction results, see Figure 5 and Figure 6 In the waterfall chart, the horizontal axis is the SHAP value; the vertical axis is the value of each feature of the sample; the red bar arrow points to the left, the SHAP value decreases, indicating that the feature has a negative impact on this prediction; the yellow bar arrow points to the right, the SHAP value increases, indicating that the feature has a positive impact on this prediction; the lower right E[f(x)] is the mean of the output values ​​of the model in the training set, that is, the baseline value of the SHAP value; from bottom to top, each feature has a corresponding positive or negative impact on the model, and the final sample SHAP value shows each feature that contributes to the model output from the baseline value to the model output value.

[0141] exist Figure 5 It can be seen that the biggest contribution to the prediction result of patient A is the patient's AST value, which is 259 and has a positive impact of 0.103 on the model; while the FDP and CA19-9 results are both 2 (exceeding the reference value range), which have a positive impact of 0.101 and 0.0991 on the model respectively; the contribution degree and direction of influence of other features to the model are explained in the same way. Starting from the baseline value E[f(x)] of the model SHAP value in the lower left corner, from bottom to top, each feature has an impact on it in different directions and degrees, and finally the f(x) of this sample (upper right) is -0.206, and the overall prediction of the occurrence of HPD in patient A is positive. Figure 6It can be seen that the most significant contribution to the prediction result of patient B is the patient's FDP, which is 1 (within the reference value range) and has a negative impact of -0.0395 on the model. The explanation of the contribution degree and influence direction of other features to the model is similar, and finally the f(x) of this sample is -0.838, and the overall prediction of the occurrence of HPD for patient B is negative.

[0142] The interpretable analysis of the prediction models for patients A and B had completely opposite results, meaning that the CAT model predicted that patient A had a very high probability of developing HPD, while patient B had a low probability of developing HPD.

[0143] Embodiment 2: In the second aspect, as Figure 7 As shown, in order to achieve the above-mentioned purpose, the present invention discloses a prediction system of a super-progression prediction model based on machine learning, comprising:

[0144] The data processing module 11 is used to obtain clinically relevant data, pre-process the clinically relevant data to obtain processed clinically relevant data, and divide the processed clinically relevant data into a training set and a test set;

[0145] The model training module 12 is used to input the training set into a pre-established Lasso regression model to screen the predictive factors, output the screened training set, and input the screened training set into a plurality of pre-established machine learning models for training to obtain a plurality of trained machine learning models;

[0146] The model prediction module 13 is used to evaluate the performance of multiple trained machine learning models according to set indicators, screen out the machine learning model with the best prediction performance, input the test set into the machine learning model with the best prediction performance, and output the prediction result.

[0147] In conjunction with the second aspect, in some implementations of the second aspect, the system further includes: the data processing module 11 preprocesses the clinical related data to perform data missing value processing, and the process is as follows:

[0148] Setting a first ratio threshold and a second ratio threshold; wherein the first ratio threshold is greater than the second ratio threshold;

[0149] Calculate the missing ratio of clinically relevant data. If the missing ratio exceeds the first ratio threshold, delete the corresponding clinically relevant data without further processing.

[0150] If the missing proportion exceeds the second proportion threshold but does not exceed the first proportion threshold, the missing clinically relevant data are uniformly marked as Unknown, and the non-missing clinically relevant data are classified according to the reference value range and converted into categorical variables in combination with the missing value group;

[0151] If the missing proportion does not exceed the second proportion threshold, the clinically relevant data are processed by the missing value interpolation method;

[0152] The Lasso regression model pre-established in the model training module 12 adds an L1 regularization term on the basis of the linear regression model, so that many parameters become zero to perform feature screening function;

[0153] The various pre-established machine learning models in the model training module 12 include a logistic regression model LR, a random forest model RF, an extreme gradient boosting tree model XGBoost, a category feature gradient boosting model CatBoost, and a support vector machine model SVM;

[0154] The indicators set in the model prediction module 12 for evaluating the performance of various trained machine learning models include area under the curve AUC, accuracy, sensitivity, specificity, positive predictive value PPV, and negative predictive value NPV;

[0155] The accuracy calculation formula in the model prediction module 13 is as follows:

[0156]

[0157] The sensitivity calculation formula is as follows:

[0158]

[0159] The specificity calculation formula is as follows:

[0160]

[0161] The positive predictive value (PPV) is calculated as follows:

[0162]

[0163] The negative predictive value (NPV) is calculated as follows:

[0164]

[0165] The area under the curve AUC is the area under the ROC curve and the coordinate axis;

[0166] Wherein, accuracy is accuracy, sensitivity is sensitivity, specificity is specificity, TP is true positive result value, TN is true negative result value, FP is false positive result value, and FN is false negative result value.

[0167] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0168] It needs to be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program is executed by a processor to execute the above method. The storage medium 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, electrical, magnetic, 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 memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, 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.

[0169] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0170] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure may have various changes and improvements, and these changes and improvements fall within the scope of the present disclosure to be protected.

Claims

1. A prediction method based on a hyperprogression prediction model of machine learning, characterized in that: The method comprises the following steps: Acquire clinically relevant data, pre-process the clinically relevant data to obtain processed clinically relevant data, and divide the processed clinically relevant data into a training set and a test set; The training set is input into a pre-established Lasso regression model to screen predictive factors, and the screened training set is outputted. The screened training set is respectively input into a plurality of pre-established machine learning models for training, and a plurality of trained machine learning models are obtained; According to the set indicators, the performance of multiple trained machine learning models is evaluated, and the machine learning model with the best prediction performance is screened out. The test set is input into the machine learning model with the best prediction performance, and the prediction result is output.

2. The prediction method of the hyperprogression prediction model based on machine learning according to claim 1, characterized in that: The clinical data is preprocessed to process missing values, and the process is as follows: Setting a first ratio threshold and a second ratio threshold; wherein the first ratio threshold is greater than the second ratio threshold; Calculate the missing ratio of clinically relevant data. If the missing ratio exceeds the first ratio threshold, delete the corresponding clinically relevant data without further processing. If the missing proportion exceeds the second proportion threshold but does not exceed the first proportion threshold, the missing clinically relevant data are uniformly marked as Unknown, and the non-missing clinically relevant data are classified according to the reference value range and converted into categorical variables in combination with the missing value group; If the missing proportion does not exceed the second proportion threshold, the clinically relevant data are processed using the missing value interpolation method.

3. The prediction method of the hyperprogression prediction model based on machine learning according to claim 1, characterized in that: The pre-established Lasso regression model adds an L1 regularization term on the basis of the linear regression model, so that many parameters become zero to perform feature screening function.

4. The prediction method of the hyperprogression prediction model based on machine learning according to claim 1, characterized in that: The multiple pre-established machine learning models include a logistic regression model LR, a random forest model RF, an extreme gradient boosting tree model XGBoost, a category feature gradient enhancement model CatBoost, and a support vector machine model SVM.

5. The prediction method of the hyperprogression prediction model based on machine learning according to claim 1, characterized in that: The set indicators for evaluating the performance of multiple trained machine learning models include area under the curve AUC, accuracy, sensitivity, specificity, positive predictive value PPV and negative predictive value NPV.

6. The prediction method of the hyperprogression prediction model based on machine learning according to claim 5, characterized in that: The accuracy calculation formula is as follows: The sensitivity calculation formula is as follows: The specificity calculation formula is as follows: The positive predictive value (PPV) is calculated as follows: The negative predictive value (NPV) is calculated as follows: The area under the curve AUC is the area under the ROC curve and the coordinate axis; Wherein, accuracy is accuracy, sensitivity is sensitivity, specificity is specificity, TP is true positive result value, TN is true negative result value, FP is false positive result value, and FN is false negative result value.

7. A prediction system based on a hyperprogression prediction model based on machine learning, characterized in that: include: A data processing module is used to obtain clinically relevant data, pre-process the clinically relevant data, obtain processed clinically relevant data, and divide the processed clinically relevant data into a training set and a test set; A model training module is used to input the training set into a pre-established Lasso regression model to screen predictive factors, output a screened training set, and input the screened training set into a plurality of pre-established machine learning models for training to obtain a plurality of trained machine learning models; The model prediction module is used to evaluate the performance of multiple trained machine learning models according to set indicators, screen out the machine learning model with the best prediction performance, input the test set into the machine learning model with the best prediction performance, and output the prediction results.

8. The prediction system of the hyperprogression prediction model based on machine learning according to claim 7, characterized in that: The data processing module preprocesses the clinical data to process missing values, and the process is as follows: Setting a first ratio threshold and a second ratio threshold; wherein the first ratio threshold is greater than the second ratio threshold; Calculate the missing ratio of clinically relevant data. If the missing ratio exceeds the first ratio threshold, delete the corresponding clinically relevant data without further processing. If the missing proportion exceeds the second proportion threshold but does not exceed the first proportion threshold, the missing clinically relevant data are uniformly marked as Unknown, and the non-missing clinically relevant data are classified according to the reference value range and converted into categorical variables in combination with the missing value group; If the missing proportion does not exceed the second proportion threshold, the clinically relevant data are processed by the missing value interpolation method; The Lasso regression model pre-established in the model training module adds an L1 regularization term to the linear regression model, making many parameters zero to perform feature screening; The model training module includes a variety of pre-built machine learning models, including logistic regression model LR, random forest model RF, extreme gradient boosting tree model XGBoost, category feature gradient boosting model CatBoost, and support vector machine model SVM; The indicators set in the model prediction module to evaluate the performance of various trained machine learning models include area under the curve AUC, accuracy, sensitivity, specificity, positive predictive value PPV and negative predictive value NPV; The accuracy calculation formula in the model prediction module is as follows: The sensitivity calculation formula is as follows: The specificity calculation formula is as follows: The positive predictive value (PPV) is calculated as follows: The negative predictive value (NPV) is calculated as follows: The area under the curve AUC is the area under the ROC curve and the coordinate axis; Wherein, accuracy is accuracy, sensitivity is sensitivity, specificity is specificity, TP is true positive result value, TN is true negative result value, FP is false positive result value, and FN is false negative result value.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: The memory stores a computer program that can be run on the processor. When the processor loads and executes the computer program, the prediction method of the hyperprogression prediction model based on machine learning described in any one of claims 1 to 6 is adopted.

10. A computer-readable storage medium having a computer program stored therein, characterized in that: When the computer program is loaded and executed by the processor, the prediction method of the hyperprogression prediction model based on machine learning described in any one of claims 1 to 6 is adopted.

Citation Information

Cited By

  • Deep learning prospecting model fair comparison evaluation method and system based on multiple constraints

    CN122064922A