Clinical features predictive of prostate cancer risk stratification - machine learning nomogram approach

Abstract

The application provides a clinical feature-machine learning nomogram method for predicting prostate cancer risk stratification, relates to data processing, data analysis, machine learning and nomogram, and constructs a clinical feature-machine learning nomogram which is easy to use, good in interpretability and powerful in function by combining machine learning and nomogram technology, supports visualization of model prediction results, and realizes the function of constructing a clinical feature-machine learning nomogram for predicting prostate cancer risk stratification. Considering the simplicity and interpretability of the nomogram and the high efficiency and robustness of the machine learning model, the clinical feature-machine learning nomogram provided by the application can be used as an auxiliary tool for preoperative evaluation of prostate cancer risk stratification, and provides necessary information for individual diagnosis and treatment of prostate cancer patients.

Description

Technical Field

[0001] This invention belongs to the field of machine learning technology, specifically relating to a machine learning nomogram method for predicting clinical features of prostate cancer risk stratification. Background Technology

[0002] Prostate-specific antigen (PSA) is a commonly used biomarker for screening and diagnosing prostate cancer in clinical practice; however, its high false-positive rate as a prostate cancer biomarker has been questioned. Minimized methods based on blood or urine (“liquid biopsy”) are increasingly being used for cancer detection, achieving precision oncology. While liquid biopsy provides clinical indicators for patients, it typically requires experienced clinicians or consultation of urology guidelines to make a comprehensive assessment of the clinical characteristics of prostate cancer patients. However, due to the subjective judgment of clinicians and the increasing number of clinical indicators, accurate risk stratification of prostate cancer patients presents challenges.

[0003] The use of machine learning algorithms helps improve the reliability and accuracy of diagnostic systems for specific diseases. Clinical decision support tools developed using machine learning algorithms can assist clinicians in quickly making a preliminary diagnosis of a patient's health condition, thereby helping them make decisions regarding treatment, medication, mental health, and other patient needs. Machine learning algorithms can automatically learn from data and extract significant features for classification tasks, and are therefore widely used in the auxiliary detection and benign / malignant analysis of prostate cancer. Machine learning methods based on peripheral blood lymphocyte subsets can distinguish between benign / malignant, or low / intermediate and high-risk prostate cancer.

[0004] In recent years, nomograms have gained popularity in clinical diagnosis due to their simplicity, intuitiveness, and interpretability. For machine learning models applied in medical auxiliary diagnosis, the interpretability of model predictions is crucial. Therefore, machine learning models built on medical clinical data often utilize nomograms to interpret their results. Based on multivariate regression, a scale score is used to characterize the changes in each variable within the multivariate regression model, and a total score is calculated to predict the probability of an event occurring. The implementation of a nomogram involves visualizing the results of logistic regression or Cox regression, establishing a scoring standard based on the magnitude of the regression coefficients of all independent variables, and assigning a score to each value level of each independent variable. A total score is calculated for each object, and then the probability of the model's predicted outcome for each object is calculated using a conversion function between the score and the probability of the outcome.

[0005] While existing machine learning methods and nomograms have made significant progress in predicting prostate cancer risk stratification, they still face several challenges. Firstly, machine learning methods do not fully meet clinical needs, exhibiting poor interpretability and limited generalizability. Secondly, nomograms are essentially constructed using regression models, whose performance is somewhat limited, resulting in poor clinical utility. To address these challenges, there is a need to develop an easy-to-use, interpretable, and powerful clinical feature—the machine learning nomogram—for non-invasive diagnosis and risk stratification assessment of prostate cancer patients. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a clinical feature-machine learning nomogram method for predicting prostate cancer risk stratification, which is used to construct a clinical feature-machine learning nomogram for predicting prostate cancer risk stratification.

[0007] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a machine learning nomogram method for predicting the clinical characteristics of prostate cancer risk stratification, comprising the following steps:

[0008] S0: Constructing machine learning models, including a machine learning model pool module, a machine learning nomogram module, and a clinical feature-machine learning nomogram module; the machine learning model pool module is used to learn from clinical data using multiple machine learning models; the machine learning nomogram module is used to construct machine learning nomograms based on the predicted probabilities of multiple machine learning models; the clinical feature-machine learning nomogram module is used to propose a feature mapping algorithm to convert machine learning nomograms into clinical feature-machine learning nomograms.

[0009] S1: The Machine Learning Model Pool module inputs a dataset consisting of multiple clinical features of an object into different machine learning models, uses the prediction results of the machine learning models to construct a machine learning nomogram, and extracts the variable scores from the machine learning nomogram.

[0010] S2: The machine learning nomogram module uses the trained machine learning model to save the corresponding feature weights for the predicted probability of each sample in the dataset and constructs a machine learning nomogram.

[0011] S3: The Clinical Feature-Machine Learning Nonograph module uses a feature mapping algorithm to combine the feature weights of clinical features in multiple machine learning models with machine learning nonograph scores to construct a clinical feature-machine learning nonograph.

[0012] S4: Obtain the score of each feature variable from the clinical feature-machine learning nomogram, and finally obtain the total label score of the object and the distribution of the corresponding low, medium and high risk stratifications.

[0013] According to the above scheme, in step S0, the machine learning model pool module includes support vector machine, decision tree, random forest, XGBoost and AdaBoost.

[0014] Furthermore, in step S1, the machine learning model is trained on a dataset composed of multiple clinical features using a 10-fold cross-validation method.

[0015] According to the above scheme, the specific steps in step S2 are as follows:

[0016] S21: The machine learning nomogram module uses the predicted probability of risk stratification for each sample in the dataset by the trained machine learning model as a new feature, saves the corresponding feature weights, and combines them with the corresponding differentiation degree label to form a new training dataset.

[0017] S22: Perform significance analysis on the new features using a logistic regression model to select effective features for constructing the nomogram;

[0018] S23: Construct a machine learning nomogram using an ordered logistic regression model based on the selected features; the variables in the machine learning nomogram are the selected machine learning models.

[0019] According to the above scheme, the specific steps in step S3 are as follows:

[0020] S31: The Clinical Features-Machine Learning Noctigraph module combines the variable scores of the machine learning noctigraph obtained in step S1 with the feature weights in the machine learning model corresponding to the variables obtained in step S2, and converts them into scores corresponding to the clinical features used to build the machine learning model through a feature mapping algorithm.

[0021] S32: Use scores to plot clinical features-machine learning nomograms; the variables in the clinical features-machine learning nomograms are the clinical features that construct the dataset.

[0022] Furthermore, in step S31, let CF i It is the score of the i-th clinical feature in the final clinical feature-machine learning nomogram; FI i,j MV is the feature weight of the i-th clinical feature in the j-th trained machine learning model; j Let be the score of the predicted probability of the j-th machine learning model in the machine learning nomogram, i∈(1,M), j∈(1,N), where M is the number of clinical features and N is the number of machine learning models; then the calculation formula of the feature mapping algorithm is:

[0023]

[0024] According to the above plan, the following steps are also included:

[0025] The performance of clinical feature-machine learning nomograms can be evaluated using ROC curves, decision curves, or calibration curves.

[0026] A computer storage medium containing a computer program executable by a computer processor, the computer program performing a clinical feature-machine learning nomogram method for predicting prostate cancer risk stratification.

[0027] The beneficial effects of this invention are as follows:

[0028] 1. The present invention provides a clinical feature-machine learning nomogram method for predicting prostate cancer risk stratification, involving data processing, data analysis, machine learning, and nomograms. By combining machine learning and nomogram technologies, it constructs an easy-to-use, interpretable, and powerful clinical feature-machine learning nomogram that supports visualization of model prediction results and realizes the function of constructing a clinical feature-machine learning nomogram for predicting prostate cancer risk stratification.

[0029] 2. Considering the simplicity and interpretability of nomograms, as well as the efficiency and robustness of machine learning models, the clinical feature-machine learning nomogram proposed in this invention can serve as an auxiliary tool for preoperative assessment of prostate cancer risk stratification, providing necessary information for individual diagnosis and treatment of prostate cancer patients. Attached Figure Description

[0030] Figure 1 This is a flowchart of the clinical feature-machine learning nomogram construction process according to an embodiment of the present invention.

[0031] Figure 2 This is a clinical feature-machine learning nomogram according to an embodiment of the present invention. Detailed Implementation

[0032] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments, using clinical characteristic data of prostate cancer as an example.

[0033] See Figure 1 The embodiments of the present invention include a machine learning model pool module, a machine learning nomogram module, and a clinical feature-machine learning nomogram module; the machine learning model pool module is used to learn from clinical data using multiple machine learning models; the machine learning nomogram module is used to construct a machine learning nomogram based on the predicted probabilities of multiple machine learning models; the clinical feature-machine learning nomogram module is used to propose a feature mapping algorithm to convert the machine learning nomogram into a new clinical feature-machine learning nomogram.

[0034] The machine learning model pool module consists of various machine learning models, including but not limited to common models such as Support Vector Machines, Decision Trees, Random Forests, XGBoost, and AdaBoost. These machine learning models are trained on a clinical dataset containing multiple features using 10-fold cross-validation. The trained model will predict the risk stratification probability for each sample in the dataset and save the corresponding feature weights.

[0035] The machine learning nomogram module is constructed using the predicted probabilities of multiple machine learning models. The predicted probability of each sample from each model is used as a new feature, combined with the corresponding risk stratification label to form a new training dataset. Then, a logistic regression model is used to perform significance analysis on the new features, selecting effective features for constructing the nomogram. Finally, based on the selected features, an ordered logistic regression model is used to construct the machine learning nomogram. The variables in this machine learning nomogram are the selected machine learning models.

[0036] The Clinical Feature-Machine Learning Nomogram module combines the variable scores from the aforementioned machine learning nomogram with the feature weights in the corresponding machine learning model, and uses a feature mapping algorithm to convert them into scores corresponding to the clinical features used to build the machine learning model. Finally, these scores are used to plot the Clinical Feature-Machine Learning Nomogram. The variables in this machine learning nomogram are the clinical features that constitute the dataset.

[0037] The calculation formula for the feature mapping algorithm is as follows:

[0038]

[0039] Among them CF i FI is the score of the i-th clinical feature in the final clinical feature-machine learning nomogram. i,j MV is the feature weight of the i-th clinical feature in the j-th trained machine learning model. j Let be the score of the j-th machine learning model prediction in the machine learning nomogram, i∈(1,M), j∈(1,N), where M is the number of clinical features and N is the number of machine learning models.

[0040] Clinical features - Machine learning nomogram final result as follows Figure 2 As shown in the figure, the score of each feature variable can be seen from the nomogram, ultimately yielding the distribution of the total label score of prostate cancer patients and their corresponding low, medium, and high risk stratifications.

[0041] The method for constructing a clinical feature-machine learning nomogram for predicting prostate cancer risk stratification according to embodiments of the present invention includes the following steps:

[0042] The first step is to train the machine learning model by inputting a dataset composed of multiple clinical features into different machine learning models, using the prediction results of these machine learning models to construct a machine learning nomogram, and extracting the machine learning nomogram score from it.

[0043] The second step is to construct a machine learning nodal plot based on the prediction results of the trained machine learning model.

[0044] The third step involves using a feature mapping algorithm to combine the feature weights of clinical features in multiple machine learning models with the machine learning nomogram scores to construct a clinical feature-machine learning nomogram.

[0045] In addition, similar to traditional nomograms, the performance of proposed clinical feature-machine learning nomograms can be evaluated using ROC curves, decision curves, or calibration curves.

[0046] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0047] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A clinical feature - machine learning nomogram method to predict prostate cancer risk stratification characterized by: The method comprises the following steps: S0: constructing a machine learning model, including a machine learning model pool module, a machine learning nomogram module and a clinical feature-machine learning nomogram module; The machine learning model pool module is used for learning clinical data using multiple machine learning models; the machine learning nomogram module is used for constructing a machine learning nomogram based on the prediction probabilities of the multiple machine learning models; and the clinical feature-machine learning nomogram module is used for proposing a feature mapping algorithm to convert the machine learning nomogram into a clinical feature-machine learning nomogram; S1: the machine learning model pool module inputs a data set composed of multiple clinical features of an object into different machine learning models, constructs a machine learning nomogram using the prediction results of the machine learning models, and extracts variable scores of the machine learning nomogram therefrom; S2: the machine learning nomogram module saves corresponding feature weights using the trained machine learning model for the prediction probability of each sample in the data set, and constructs a machine learning nomogram; S3: the clinical feature-machine learning nomogram module constructs a clinical feature-machine learning nomogram by a feature mapping algorithm, in combination with the feature weights of the clinical features in the multiple machine learning models and the machine learning nomogram scores; S4: the scores of each feature variable are obtained from the clinical feature-machine learning nomogram, and finally the total label score of the object and the distribution of the corresponding low, medium and high risk stratifications are obtained.

2. The clinical features-machine learning nomogram method of predicting prostate cancer risk stratification of claim 1, wherein: In the step S0, the machine learning model pool module comprises a support vector machine, a decision tree, a random forest, XGBoost and AdaBoost.

3. The clinical features-machine learning nomogram method of predicting prostate cancer risk stratification of claim 2, wherein: In the step S1, the machine learning model is trained in a data set composed of multiple clinical features by using a 10-fold cross-validation method.

4. The clinical features-machine learning nomogram method of predicting prostate cancer risk stratification of claim 1, wherein: In the step S2, the specific steps are as follows: S21: the machine learning nomogram module uses the trained machine learning model to save the corresponding feature weights using the prediction probability of the risk stratification of each sample in the data set as a new feature, and combines the corresponding differentiation degree label to form a new training data set; S22: significant analysis is performed on the new feature by using a logistic regression model to screen effective features for constructing a nomogram; S23: an ordered logistic regression model is used to construct a machine learning nomogram based on the screened features; the variable of the machine learning nomogram is the screened machine learning model.

5. The clinical features-machine learning nomogram method of predicting prostate cancer risk stratification of claim 1, wherein: In the step S3, the specific steps are as follows: S31: the clinical feature-machine learning nomogram module converts the variable score of the machine learning nomogram obtained in the step S1 into a score corresponding to the clinical feature for constructing the machine learning model by a feature mapping algorithm in combination with the feature weight of the variable in the machine learning model obtained in the step S2; S32: the score is used to draw a clinical feature-machine learning nomogram; the variable of the clinical feature-machine learning nomogram is the clinical feature for constructing the data set.

6. The clinical features-machine learning nomogram method of predicting prostate cancer risk stratification of claim 5, characterized in that: The step S31 is provided with CF i is the score of the i-th clinical feature in the final obtained clinical feature-machine learning nomogram; FI i,j is the feature weight of the i-th clinical feature in the j-th trained machine learning model; MV j is the score of the jth machine learning model prediction probability in the machine learning nomogram, i∈(1, M), j∈(1, N), where M is the number of clinical features, and N is the number of machine learning models; and the calculation formula of the feature mapping algorithm is:

7. The clinical features-machine learning nomogram method of predicting prostate cancer risk stratification of claim 1, wherein: The method further comprises the following steps: The performance of the clinical feature-machine learning nomogram is evaluated by using a ROC curve, a decision curve or a calibration curve.

8. A computer storage medium, characterized in that: having computer readable program code stored therein, which computer readable program code is executable by a computer processor to perform the clinical feature-machine learning nomogram method of predicting prostate cancer risk stratification as claimed in any one of claims 1 to 7.

Images