Colorectal cancer postoperative venous thromboembolism risk prediction method and system

Abstract

The invention discloses a colorectal cancer postoperative venous thromboembolism risk prediction method and system, and the method comprises the steps: obtaining clinical pathology data of a historical patient before the patient receives a total proctectomy, carrying out the preprocessing of the clinical pathology data, employing a standardized score to detect an abnormal value in the clinical pathology data, carrying out the filling of a missing value, and carrying out the prediction of the risk of the venous thromboembolism. Performing standardization processing on the continuous variables; based on the preprocessed clinical pathology data, extracting clinical pathology features by adopting Lasso regression and grid search; constructing a Stacking integration model based on a random forest, a CatBoost model and logistic regression, and training the Stacking integration model by using the extracted clinical pathological features; and by using the trained Stacking integrated model, processing clinical pathology data of a target patient before the target patient receives the total proctectomy to obtain a prediction result. The colorectal cancer postoperative venous thromboembolism risk can be effectively predicted, and support is provided for clinical decision making.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical big data analysis, and relates to a method and system for predicting the risk of venous thromboembolism after colorectal cancer surgery, and in particular to a method and system for predicting the risk of venous thromboembolism after colorectal cancer surgery based on clinical pathological data. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] The current standard surgical option for colorectal cancer in clinical practice is total mesorectal excision. During the perioperative treatment of colorectal cancer, there are often problems with venous thrombosis, the most common of which is venous thromboembolism (VTE). Venous thromboembolism (VTE) is a common and serious complication after colorectal cancer surgery, including deep vein thrombosis (DVT) and pulmonary embolism (PE). Venous thromboembolism can cause prolonged hospitalization, post-thrombotic syndrome, significant increase in medical expenses, and even death.

[0004] As a multifactorial disease, venous thromboembolism has the characteristics of hidden clinical manifestations, heterogeneous clinical phenotypes, mixed risk factors, difficult treatment decisions, and high incidence of complications. In actual clinical treatment, doctors cannot predict whether patients will develop venous thromboembolism after surgery, which leads to poor prognosis for some patients.

[0005] With the help of machine learning models and big data analysis technology, it is possible to mine features related to postoperative venous thromboembolism from preoperative pathological data, and establish an accurate prediction model to quantitatively assess the risk of complications of venous thromboembolism. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a method and system for predicting the risk of venous thromboembolism after colorectal cancer surgery. The present invention can mine the mapping relationship between preoperative pathological data and postoperative complications of venous thromboembolism, thereby realizing the prediction of the risk of venous thromboembolism.

[0007] According to some implementation examples, the present invention adopts the following technical solutions:

[0008] A method for predicting the risk of venous thromboembolism after colorectal cancer surgery comprises the following steps:

[0009] Obtain clinical pathological data of historical patients before total mesorectal excision;

[0010] Preprocessing the clinical pathology data, including detecting abnormal values ​​therein using standardized scores, filling missing values, and standardizing continuous variables;

[0011] Based on the preprocessed clinical pathological data, Lasso regression and grid search were used to extract clinical pathological features;

[0012] Building a Stacking ensemble model based on random forest, CatBoost model and logistic regression, and training the Stacking ensemble model using the extracted clinical pathological features;

[0013] The trained Stacking ensemble model was used to process the clinical pathological data of the target patients before total mesorectal excision to obtain the prediction results.

[0014] As an optional embodiment, when obtaining the clinical pathological data of historical patients before undergoing total mesorectal excision, corresponding labels are assigned for venous thromboembolism, including venous thromboembolism positive and venous thromboembolism negative.

[0015] As an optional implementation, the process of using the standardized score to detect outliers includes:

[0016] Define the Contiuous_OutlierCheck() function, read the relevant files from the specified path, and load them into a data frame;

[0017] Select the columns of continuous variables and calculate the Z-Score for the data in each column:

[0018]

[0019] Among them, Z represents the standardized score, X represents the original data point, μ represents the mean of the column data, and σ represents the standard deviation of the column data. According to the statistical definition, the data point with |Z|>3 is judged as an outlier.

[0020] As an optional implementation, the missing values ​​are filled as follows:

[0021] For the ordered categorical variables in the pathological data, the OrdinalEncoder method in the sklearn-preprocesing library is used to perform ordinal encoding on the variables, and the ordered categorical variables include clinical T stage and N stage;

[0022] For the unordered categorical variables in the pathological data, the unordered categorical variables include whether there is postoperative bleeding, whether there is a history of smoking, whether there is a history of drinking, and the surgical method;

[0023] For binary data in pathological data, use the LabelEncoder method for label encoding;

[0024] For multi-classification data in pathological data, the OneHotEncoder method is used for one-hot encoding;

[0025] Randomly fill the data in the randomly selected columns of the categorical variables in the pathological data;

[0026] For the missing values ​​of continuous variables in the pathological data, the IterativeImputer method in the sklearn-impute library was used to fit the missing data using multiple imputation of the Bayesian Ridge regression model.

[0027] As an optional implementation, the extraction of clinical pathological features is specifically as follows:

[0028] Use the LassoCV function of the sklearn.linear_model library in the Lasso regression method for feature selection, perform cross-validation among multiple regularization parameters, and automatically select the optimal regularization parameter.

[0029] As an optional implementation, the construction of the Stacking integration model is specifically as follows:

[0030] Initialize the base learners Random Forest and CatBoost;

[0031] Initialize the meta-learner logistic regression model;

[0032] Using 4-fold cross validation, the training data is randomly divided into 4 equal parts to form 4 sets of data sets, of which 3 / 4 of each set of data sets is the training set and the other 1 / 4 is the validation set;

[0033] The training set of each group is input into the Random Forest and CatBoost models for training, and then the validation set is predicted to obtain the prediction probability of the corresponding validation set;

[0034] Stack the prediction results of a base learner for each validation set row by row to obtain the prediction value of the complete sample data;

[0035] The complete sample prediction results of the two base learners are concatenated by column to form new sample data, and the new sample data is input into the meta learner for fitting;

[0036] The BayesSearchCV method of the sklearn.ensemble library is used to optimize the parameter selection of the Stacking ensemble model so that the model achieves the optimal AUC indicator on the test set.

[0037] As an optional implementation, the Stacking integration model is specifically:

[0038] It consists of Random Forest, CatBoost base learners and logistic regression meta-learner. It uses 4-fold cross validation and Bayesian search algorithm to optimize model parameters so that the AUC of the model on the test set reaches the optimal value.

[0039] A system for predicting the risk of venous thromboembolism after colorectal cancer surgery, comprising:

[0040] A data acquisition module is configured to acquire clinical pathological data of historical patients before undergoing total mesorectal excision;

[0041] A data preprocessing module is configured to preprocess the clinical pathology data, wherein the preprocessing includes detecting abnormal values ​​therein using standardized scores, filling missing values, and standardizing continuous variables;

[0042] The feature extraction module is configured to extract clinical pathological features based on the preprocessed clinical pathological data using Lasso regression and grid search;

[0043] A model building module is configured to build a Stacking ensemble model based on random forest, CatBoost model and logistic regression, and train the Stacking ensemble model using the extracted clinical pathological features;

[0044] The risk prediction module is configured to use the trained Stacking ensemble model to process the clinical pathological data of the target patient before total mesorectal excision to obtain a prediction result.

[0045] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the steps in the above method are completed.

[0046] An electronic device comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the above method are completed.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] (1) The present invention utilizes the preoperative pathological examination data of colorectal cancer patients and uses machine learning technology to mine the features that can be used to predict the onset of postoperative venous thromboembolism complications, thereby improving the prediction accuracy of venous thromboembolism onset and enhancing patients' risk awareness.

[0049] (2) The present invention makes full use of clinical data and big data analysis tools, and the important features screened out by Lasso and grid search can be used for subsequent correlation studies between diseases and preoperative characteristics.

[0050] (3) The present invention utilizes the idea of ​​Stacking to construct a machine learning model with powerful fitting capabilities, and combines the Bayesian search algorithm and K-fold cross-validation to enhance the reliability of the model's prediction performance.

[0051] Other features and advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of a method for predicting the risk of venous thromboembolism after colorectal cancer surgery based on clinical pathological data in an embodiment of the present invention;

[0053] Figure 2 It is a Lasso regression path diagram in an embodiment of the present invention;

[0054] Figure 3 A schematic diagram of a Stacking integrated model training pipeline in an embodiment of the present invention;

[0055] Figure 4 These are the features and their weights selected in the embodiment of the present invention. DETAILED DESCRIPTION

[0056] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0057] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0058] Embodiment 1

[0059] In one or more embodiments, a method for predicting the risk of venous thromboembolism after colorectal cancer surgery based on clinical pathological data is disclosed, combined with Figure 1 , specifically including the following process:

[0060] S101: Obtain the clinical pathological data of the patient before total mesorectal excision;

[0061] S102: Data preprocessing using methods such as data cleaning, missing value filling, variable coding, and standardization;

[0062] Among them, the Z-score standardized score is used to clean up erroneous data, specifically:

[0063] Define the Contiuous_OutlierCheck() function, read the Excel file from the specified path, and load it into a data frame. Select the columns of continuous variables and calculate the Z-Score for the data in each column:

[0064]

[0065] Among them, Z represents the standardized score, X represents the original data point, μ represents the column data mean, and σ represents the column data standard deviation. In statistics, data points with |Z|>3 are judged as outliers and are set to NAN;

[0066] Variable encoding and missing value filling, specifically:

[0067] For ordered categorical variables in clinical pathology data, including clinical T stage, N stage, etc., the OrdinalEncoder method in the sklearn-preprocesing library is used to ordinal encode the variables. The ordinal encoder converts the categories into integers (starting from 0) according to the input order during mapping. When there is an order in the category during the encoding process (such as cancer stages from I to IV), this encoding method will minimize information loss; for unordered categorical variables, including whether there is postoperative bleeding, whether there is a history of smoking, whether there is a history of drinking, and surgical methods, etc.; the binary data uses the LabelEncoder method for label encoding to convert the categorical variables into numerical data that the model can handle; the multi-category data uses the OneHotEncoder method for one-hot encoding to avoid the assumed order between categories. After encoding, each eigenvalue of the categorical variable will correspond to a new binary column; for the missing values ​​of the categorical variable, the data in the column is randomly selected to fill; for the missing values ​​of the continuous variable, the IterativeImputer method in the sklearn-impute library is used to fit the missing data using the BayesianRidge model;

[0068] S103: Use Lasso and grid search to screen important features;

[0069] Among them, Lasso screening features are specifically:

[0070] Lasso (Least Absolute Shrinkage and Selection Operator) is a widely used regression technique that can reduce the coefficients of unimportant features to zero by introducing L1 regularization, thereby achieving feature selection;

[0071] In this example, we used LassoCV for feature selection. LassoCV is an automated Lasso regression method that can perform cross-validation in multiple regularization parameters (i.e., alpha values) and automatically select the optimal alpha. These values ​​are between 10 -4 -10 1 The model is trained on the training data by 5-fold cross validation (cv=5). In each cross validation, the data is divided into a training set and a validation set. The model is trained on the training set and the performance is evaluated on the validation set. Finally, the alpha value with the best performance is selected (combined with Figure 2 and Figure 3 , the optimal alpha in this case is 0.0093, and 15 relevant features are selected, such as Figure 4 As shown, specifically including total bilirubin (TBil), platelet Plt, postoperative infection SSI, cancer marker Ca, albumin Albumin, nutritional risk screening NRS2002, intracranial atherosclerotic stenosis lcas, activated partial thromboplastin time APTT, low-density lipoprotein LDL, length of stay, surgical method Operation method, D-dimer D-dimer, acute renal failure ARF, surgical site Surgical site).

[0072] S104: Input the selected features into the Stacking ensemble model for training, and verify the model performance on the test set;

[0073] Combination Figure 4 , the base learners are initialized as Random Forest and CatBoost models, and the meta learner is initialized as the logistic regression model;

[0074] As a specific implementation method, firstly, a training data set is constructed to train the MRI-endoscopy multimodal fusion model; the specific process is as follows:

[0075] (1) Based on the features selected by Lasso, a new sample matrix is ​​constructed:

[0076]

[0077] Where N = 1564 represents the number of samples, and M = 15 represents the number of features;

[0078] (2) Construct the Stacking integration model. The specific process is as follows:

[0079] Instantiate the StackingClassifier object in the sklearn.ensemble library; initialize the base learner: estimators = [('rf', RandomForestClassifier()), ('catboost', CatBoostClassifier())]; initialize the meta learner: final_estimator = LogisticRegression(); Bayesian search optimization parameters,

[0080] Define the hyperparameter search space:

[0081] 'rf__n_estimators':(10,200),

[0082] 'rf__max_depth':(3,10),

[0083] 'catboost__depth':(4,10),

[0084] 'catboost__learning_rate':(0.001,0.3),

[0085] 'catboost__iterations':(100,10000),

[0086] 'catboost__l2_leaf_reg':(1,10),

[0087] The model was trained in 4-fold cross validation using the BayesSearchCV() function.

[0088] Embodiment 2

[0089] In one or more embodiments, a system for predicting the risk of venous thromboembolism after colorectal cancer surgery is disclosed, comprising:

[0090] A data acquisition module is configured to acquire clinical pathological data of historical patients before undergoing total mesorectal excision;

[0091] A data preprocessing module is configured to preprocess the clinical pathology data, wherein the preprocessing includes detecting abnormal values ​​therein using standardized scores, filling missing values, and standardizing continuous variables;

[0092] The feature extraction module is configured to extract clinical pathological features based on the preprocessed clinical pathological data using Lasso regression and grid search;

[0093] A model building module is configured to build a Stacking ensemble model based on random forest, CatBoost model and logistic regression, and train the Stacking ensemble model using the extracted clinical pathological features;

[0094] The risk prediction module is configured to use the trained Stacking ensemble model to process the clinical pathological data of the target patient before total mesorectal excision to obtain a prediction result.

[0095] The Stacking ensemble model consists of Random Forest, CatBoost base learners and logistic regression meta-learner, and uses 4-fold cross validation and Bayesian search algorithm to optimize model parameters so that the AUC of the model on the test set reaches the optimal value.

[0096] The specific implementation of the above modules is the same as that in Example 1 and will not be described in detail.

[0097] Embodiment 3

[0098] In one or more embodiments, a terminal device is disclosed, including a server, the server including a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor implementing the method for predicting the risk of postoperative venous thromboembolism after colorectal cancer surgery based on clinical pathological data in Example 1 when executing the program. For the sake of brevity, it will not be described in detail here.

[0099] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0100] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0101] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software.

[0102] Embodiment 4

[0103] In one or more embodiments, a computer-readable storage medium is disclosed, in which a plurality of instructions are stored, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the method for predicting the risk of venous thromboembolism after colorectal cancer surgery based on clinical pathological data as described in Example 1.

[0104] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for predicting the risk of venous thromboembolism after colorectal cancer surgery, characterized in that: The following steps are involved: Obtain clinical pathological data of historical patients before total mesorectal excision; Preprocessing the clinical pathology data, including detecting abnormal values ​​therein using standardized scores, filling missing values, and standardizing continuous variables; Based on the preprocessed clinical pathological data, Lasso regression and grid search were used to extract clinical pathological features; Building a Stacking ensemble model based on random forest, CatBoost model and logistic regression, and training the Stacking ensemble model using the extracted clinical pathological features; The trained Stacking ensemble model was used to process the clinical pathological data of the target patients before total mesorectal excision to obtain the prediction results.

2. A method for predicting the risk of venous thromboembolism after colorectal cancer surgery as claimed in claim 1, characterized in that: When obtaining the clinical pathological data of historical patients before undergoing total mesorectal excision, corresponding labels were assigned for venous thromboembolism, including venous thromboembolism positive and venous thromboembolism negative.

3. The method for predicting the risk of venous thromboembolism after colorectal cancer surgery according to claim 1, characterized in that: The process of using the standardized score to detect outliers includes: Define the Contiuous_OutlierCheck() function, read the relevant files from the specified path, and load them into a data frame; Select the columns of continuous variables and calculate the Z-Score for the data in each column: Among them, Z represents the standardized score, X represents the original data point, μ represents the mean of the column data, and σ represents the standard deviation of the column data. According to the statistical definition, the data point with |Z|>3 is judged as an outlier.

4. The method for predicting the risk of venous thromboembolism after colorectal cancer surgery according to claim 1, characterized in that: The missing values ​​are filled as follows: For the ordered categorical variables in the pathological data, the OrdinalEncoder method in the sklearn-preprocesing library is used to perform ordinal encoding on the variables, and the ordered categorical variables include clinical T stage and N stage; For the unordered categorical variables in the pathological data, the unordered categorical variables include whether there is postoperative bleeding, whether there is a history of smoking, whether there is a history of drinking, and the surgical method; For binary data in pathological data, use the LabelEncoder method for label encoding; For multi-classification data in pathological data, the OneHotEncoder method is used for one-hot encoding; Randomly fill the data in the randomly selected columns of the categorical variables in the pathological data; For the missing values ​​of continuous variables in the pathological data, the IterativeImputer method in the sklearn-impute library was used to fit the missing data using multiple imputation of the Bayesian Ridge regression model.

5. The method for predicting the risk of venous thromboembolism after colorectal cancer surgery according to claim 1, characterized in that: The extraction of clinical pathological features is specifically as follows: Use the LassoCV function of the sklearn.linear_model library in the Lasso regression method for feature selection, perform cross-validation among multiple regularization parameters, and automatically select the optimal regularization parameter.

6. The method for predicting the risk of venous thromboembolism after colorectal cancer surgery according to claim 1, characterized in that: The construction of the Stacking integration model is specifically as follows: Initialize the base learners Random Forest and CatBoost; Initialize the meta-learner logistic regression model; Using 4-fold cross validation, the training data is randomly divided into 4 equal parts to form 4 sets of data sets, of which 3 / 4 of each set of data sets is the training set and the other 1 / 4 is the validation set; The training set of each group is input into the Random Forest and CatBoost models for training, and then the validation set is predicted to obtain the prediction probability of the corresponding validation set; Stack the prediction results of a base learner for each validation set row by row to obtain the prediction value of the complete sample data; The complete sample prediction results of the two base learners are concatenated by column to form new sample data, and the new sample data is input into the meta learner for fitting; The BayesSearchCV method of the sklearn.ensemble library is used to optimize the parameter selection of the Stacking ensemble model so that the model achieves the optimal AUC indicator on the test set.

7. The method for predicting the risk of venous thromboembolism after colorectal cancer surgery according to claim 1, characterized in that: The Stacking integration model is specifically: It consists of Random Forest, CatBoost base learners and logistic regression meta-learner. It uses 4-fold cross validation and Bayesian search algorithm to optimize model parameters so that the AUC of the model on the test set reaches the optimal value.

8. A system for predicting the risk of venous thromboembolism after colorectal cancer surgery, characterized in that: include: A data acquisition module, configured to acquire clinical pathological data of historical patients before undergoing total mesorectal excision; A data preprocessing module is configured to preprocess the clinical pathology data, wherein the preprocessing includes detecting abnormal values ​​therein using standardized scores, filling missing values, and standardizing continuous variables; The feature extraction module is configured to extract clinical pathological features based on the preprocessed clinical pathological data using Lasso regression and grid search; A model building module is configured to build a Stacking ensemble model based on random forest, CatBoost model and logistic regression, and train the Stacking ensemble model using the extracted clinical pathological features; The risk prediction module is configured to use the trained Stacking ensemble model to process the clinical pathological data of the target patient before total mesorectal excision to obtain a prediction result.

9. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps in the method according to any one of claims 1 to 7 are completed.

Images