System for predicting survival rate of patient after lung cancer operation
By constructing a lung cancer postoperative survival prediction system based on logistic regression algorithm, and utilizing simplified decision functions and gradient ascent method to iterate parameters, combined with data preprocessing, a high-accuracy prediction of postoperative survival rate for lung cancer patients was achieved. This solves the problem of low prediction accuracy in existing technologies and improves the accuracy and efficiency of medical diagnosis.
Patent Information
- Application Number
- CN202411100620.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
There is a lack of existing technologies for predicting the survival rate of lung cancer patients after surgery based on logistic regression algorithms, and the disease prediction accuracy of existing technologies based on logistic regression algorithms is low, which urgently needs to be improved.
A system for predicting the survival rate of lung cancer patients after surgery is constructed, including modules for data collection, data preprocessing, prediction and evaluation, and result output. By simplifying the logistic regression decision function, the parameters are determined iteratively using the gradient ascent method. Combined with discrete feature encoding and standardization, a prediction model for assessing the survival rate of lung cancer patients after surgery is constructed.
It achieves a high accuracy rate in predicting the postoperative survival rate of lung cancer patients, with an accuracy rate of over 80% and a prediction time of 0.2 seconds, significantly improving the accuracy and efficiency of traditional medical diagnosis.
Smart Images

Figure CN119008008A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of bioinformatics, and in particular relates to a system for predicting the survival rate of patients after lung cancer surgery. Background Art
[0002] Lung cancer is a malignant tumor of the bronchial mucosa or glands of the lungs, which poses a great threat to people's health and life, especially because of the decline in air quality and the increasing frequency of smog and other weather conditions, which seriously harm people's respiratory organs and cause lung diseases. On the other hand, due to people's unhealthy habits, such as smoking, alcoholism, and staying up late, the incidence and mortality of lung cancer have increased significantly. The increase in the number of lung cancer patients has made medical clinical data sets more complex, and it has become more difficult to diagnose the disease only through manual extraction and integration. Therefore, it is imperative to introduce a new prediction method to assist manual disease diagnosis.
[0003] Machine learning is an important branch of artificial intelligence, involving probability theory, statistics, etc. It is an important means for artificial intelligence to simulate human learning. Machine learning is mainly divided into two aspects. The first is "training", that is, building a model through various existing learning materials. The more learning materials, the more accurate the model building. The second is "decision-making", which is to use the model built in the first step of training for analysis and calculation. The five commonly used machine learning algorithm models include naive Bayes model, logistic regression model, support vector machine model, decision tree model and random forest model.
[0004] Logistic regression is a classic method in machine learning. It is computationally inexpensive and easy to understand and implement. The logistic regression algorithm is a typical supervised learning classification method. The algorithm model is represented by conditional probability, which combines the sigmoid function and the linear superposition function. The decision function is: Ρ<γ=0│X>=1 / (1+e^(w·x+b)), Ρ<γ=1│X>=e^(w·x+b) / (1+e^(w·x+b)). The principle of the algorithm is to use the maximum log-likelihood function to obtain the best probability model for the training set samples through supervised learning, and finally use the model to analyze and predict the new sample data. Chinese patent CN108492884A discloses a system for predicting pancreatic neuroendocrine tumor lymph node metastasis based on a Logistic regression model, the system comprising a memory, an information processing module, and a prediction result display, wherein the information processing module is configured to perform the following calculations, and compare the calculation results with the pancreatic neuroendocrine tumor lymph node metastasis prediction indicators pre-stored in the memory, and output the prediction results to the prediction result display: (I) Calculate the scores of each risk factor according to the following relationship: (II) Sum all the scores obtained in step (I) to calculate the overall score Z; (III) Calculate the possibility of lymph node metastasis of the patient according to the following relationship. The prediction accuracy of the system is 0.646.
[0005] However, the current existing technology lacks a system for predicting the survival rate of patients after lung cancer surgery based on the logistic regression algorithm, and the accuracy of disease prediction based on the logistic regression algorithm in the existing technology is low. There is an urgent need for a new logistic regression model prediction system with significantly improved prediction accuracy. Summary of the invention
[0006] As used herein, the singular forms "a," "an," and "the" include both singular and plural referents unless the context clearly dictates otherwise. Numerical ranges recited by endpoints include all numbers and fractions within the corresponding ranges, as well as the recited endpoints.
[0007] In view of the problems existing in the prior art, the present invention provides a system for predicting the survival rate of patients after lung cancer surgery. The system can predict and judge the survival rate of lung cancer patients one year after lung resection, and has high prediction accuracy and efficiency, and can effectively assist medical diagnosis.
[0008] To achieve the above purpose, the technical solution adopted by the present invention is as follows:
[0009] A system for predicting the survival rate of patients after lung cancer surgery, comprising a data collection module, a data preprocessing module, a prediction evaluation module and a result output module;
[0010] The prediction and evaluation module includes a prediction model for evaluating the survival rate of lung cancer patients after surgery; the method for constructing the prediction model for evaluating the survival rate of lung cancer patients after surgery includes the following steps:
[0011] (1) Simplify the decision function of logistic regression and obtain a simplified decision function:
[0012] The simplified decision function is:
[0013]
[0014] (2) Using the simplified decision function to estimate the parameters, we can obtain the log-likelihood function:
[0015] The log-likelihood function is
[0016]
[0017] (3) The log-likelihood function is iterated using the gradient ascent method to obtain the maximum value of the objective function L(w). At this time, the logistic regression model determined by w is the prediction model for evaluating the postoperative survival rate of lung cancer patients.
[0018] Preferably, the gradient direction of the gradient in step (3) is
[0019]
[0020] Preferably, the iteration in step (3) is: iterating the parameter w in the gradient direction, specifically:
[0021] Preferably, the data collection module is used to obtain samples of lung cancer patients who undergo lung resection surgery; the lung cancer patient samples include at least 7 of the following: type of lung cancer patient tumor, forced vital capacity, volume of air exhaled in the first second of maximum exhalation, Zhu score, pre-operative pain, pre-operative hemoptysis, pre-operative dyspnea, pre-operative cough, pre-operative weakness, original tumor size, type 2 diabetes, peripheral arterial disease, smoking status, asthma status and patient age; preferably at least 10.
[0022] Preferably, the data preprocessing module includes discrete feature encoding and standardization processing of the data.
[0023] The formula for the standardization process is:
[0024]
[0025] Among them, x represents the original eigenvalue, μ is the mean of the column, σ is the standard deviation of the column, and x* is the final eigenvalue.
[0026] Preferably, the data of the data preprocessing module includes training set data for training a prediction model for evaluating the postoperative survival rate of lung cancer patients.
[0027] The training method is: substituting the training set data into the obtained lung cancer patient postoperative survival rate assessment prediction model, calculating the probabilities p(y=1|x) and p(y=0|x), and comparing the two probabilities, and taking the value of the item y with a larger probability as the category of the new sample.
[0028] Preferably, the result output module is used to output the survival rate of lung cancer patients one year after lung resection.
[0029] Preferably, the method for predicting the survival rate of patients after lung cancer surgery by the system comprises the following steps:
[0030] S1. Use the data collection module to obtain the sample data;
[0031] S2, using a data preprocessing module to preprocess the data set selected in step S1;
[0032] S3, inputting the data pre-processed in step S2 into the prediction and evaluation module;
[0033] S4. Obtain an evaluation report on the survival rate of patients after lung cancer surgery through the result output module.
[0034] Compared with the prior art, the present invention has the following beneficial effects:
[0035] The present invention provides a system for predicting the survival rate of patients after lung cancer surgery. The system is based on a logistic regression model and can predict and judge the survival rate of lung cancer patients one year after lung resection. Compared with traditional medical diagnosis, the prediction accuracy and efficiency of the system are greatly improved, with an accuracy rate of more than 80%, and the prediction time is only 0.2 seconds, which can most effectively assist medical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 Flow chart of the system of the present invention. DETAILED DESCRIPTION
[0037] The present invention is described below through specific embodiments to make the technical solution of the present invention easier to understand and grasp, but the present invention is not limited thereto, and the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.
[0038] The endpoints and any values of the ranges disclosed in this article are not limited to the precise ranges or values, and these ranges or values should be understood to include values close to these ranges or values. For numerical ranges, the endpoint values of each range, the endpoint values of each range and the individual point values, and the individual point values can be combined with each other to obtain one or more new numerical ranges, which should be considered as specifically disclosed in this article.
[0039] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application. The experimental methods described in the following embodiments are conventional methods unless otherwise specified; the reagents and materials described are commercially available unless otherwise specified.
[0040] Example 1 Construction of a prediction model for evaluating the survival rate of lung cancer patients after surgery
[0041] S1. We selected 470 samples of lung cancer patients who underwent lung resection at the Wroclaw Thoracic Surgery Center from 2007 to 2011. Through a large number of experiments, we determined that the collected data features include 16 features such as patients' lung-related indicators and preoperative manifestations. The specific features are shown in Table 1. The goal of the data set is to predict whether the patient can survive one year after surgery. The label of the survival data is 1, and the label of the non-survival data is 0.
[0042] Table 1 Dataset feature fields and feature explanations
[0043] Feature Name Feature explanation 1.DGN Lung cancer patient tumor type 2.PRE4 forced vital capacity 3.PRE5 The volume of air exhaled in the first second of maximum exhalation 4.PRE6 Zhu's score (PRZ2, PRZ1, PRZ0) 5.PRE7 Whether there was pain before surgery (T, F) 6.PRE8 Whether there is hemoptysis before surgery (T, F) 7.PRE9 Whether there is dyspnea before surgery (T, F) 8.PRE10 Whether there was cough before surgery (T, F) 9.PRE11 Whether the patient was weak before surgery (T, F) 10.PRE14 Original tumor size (small to large OC11, OC14, OC12, OC13) 11.PRE17 Whether there is type 2 diabetes (T, F) 12.PRE19 MI up to 6 months (T, F) 13.PRE25 Whether there is peripheral arterial disease (T, F) 14.PRE30 Do you smoke (T, F) 15.PRE32 Whether asthma (T, F) 16.age Patient age at surgery
[0044] S2. Data preprocessing:
[0045] In order to avoid irregularities and inconsistencies in the training data, the data selected in step S1 needs to be preprocessed before model training to convert the original data set into a set with a unified interval and the same dimension that can be recognized by the machine. The data preprocessing method used is discrete feature encoding and standardization.
[0046] (1) Discrete feature encoding
[0047] The discrete feature encoding method used is one-hot encoding. All features containing more than 2 discrete feature values need to be modified into one-hot encoding. For example, the tumor type feature of lung cancer patients contains 7 feature values. One-hot encoding converts the feature into 7 sub-features. For example, the tumor type dgn1 is encoded as 1000000 and dgn2 is encoded as 0100000. If one-hot encoding is not performed, the concept of digital "sequence" will be introduced. Through one-hot encoding, all features will be fair and unified for machine learning.
[0048] (2) Standardization
[0049] The purpose of standardization is to eliminate the impact of different feature data dimensions and numerical values, so as to ensure that the construction of machine learning models will not be affected by the characteristics of the data itself. The standardization formula is as follows:
[0050]
[0051] Where x represents the original eigenvalue, μ is the mean of the column, σ is the standard deviation of the column, and x* is the final eigenvalue. The result of standardization is to make all eigenvalues of each feature obey the standard normal distribution.
[0052] S3. Construct a prediction model for evaluating the postoperative survival rate of lung cancer patients
[0053] 1. Simplify the decision function:
[0054] For decision function Expand the feature vector X and the weight vector W to remove the parameter b in the model. That is, change X to (X (1) ,X (2) ,…,X (n) ,1),W is changed to
[0055] (w (1) ,w (2) ,…,w (n) ,b).
[0056] At this time, the decision function of logistic regression is simplified to:
[0057]
[0058]
[0059] 2. Obtain the log-likelihood function:
[0060] First, let P<γ=1|χ>=π(χ) and P<γ=0|χ>=1-π(χ).
[0061] So the likelihood function is:
[0062]
[0063] So the log-likelihood function is:
[0064]
[0065] 3. Construct a prediction model for evaluating the survival rate of patients with lung cancer after surgery:
[0066] Then, through the gradient ascent method, the value of variable w is continuously changed to make the log-likelihood function L(w) larger. When the function L(w) reaches its maximum value, the variable w is obtained and w is brought into If you hit the mark, you’ll get it.
[0067] Among them, the method for solving the gradient direction of the gradient is: find the partial derivative of the objective function with respect to the variable.
[0068] The objective function is:
[0069]
[0070] So the gradient direction is:
[0071]
[0072] According to the definition of gradient, when the variable w of the objective function L(w) moves along the gradient direction, the objective function grows
[0073] The fastest. So we need to iterate w in the direction of the gradient, that is:
[0074]
[0075] Until the final objective function reaches a maximum value, the parameter w can determine a logistic regression model with the best classification effect, which is the prediction model for the postoperative survival rate of lung cancer patients.
[0076] S4. Model Evaluation
[0077] The training set data is divided into a training set and a test set, where the training set is used to train the model and the test set is used to predict the model quality. The model evaluation indicator is the accuracy rate, which is the proportion of correctly classified samples to the total samples, that is:
[0078]
[0079] The accuracy of the model constructed by the present invention is stable at 70% to 80%, and the running time is about 0.2 seconds. The results show that the present invention has higher accuracy and efficiency in predicting results by collecting specific data and using specific preprocessing methods and prediction models.
[0080] Example 2: Method for predicting the survival rate of patients after lung cancer surgery using a system for predicting the survival rate of patients after lung cancer surgery
[0081] S1. Use the data collection module to obtain the sample data; the specific sample data is shown in Table 1.
[0082] S2, using a data preprocessing module to preprocess the data set selected in step S1;
[0083] S3, inputting the data pre-processed in step S2 into the prediction and evaluation module;
[0084] S4. Obtain an evaluation report on the survival rate of patients after lung cancer surgery through the result output module.
[0085] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions of the technical solution of the present invention by ordinary technicians in this field do not deviate from the essence and scope of the technical solution of the present invention.
Claims
1. A system for predicting the survival rate of patients after lung cancer surgery. Features: It includes data collection module, data preprocessing module, prediction and evaluation module and result output module; The prediction and evaluation module includes a prediction model for evaluating the survival rate of lung cancer patients after surgery; the method for constructing the prediction model for evaluating the survival rate of lung cancer patients after surgery includes the following steps: (1) Simplify the decision function of logistic regression and obtain a simplified decision function: The simplified decision function is: (2) Using the simplified decision function to estimate the parameters, we can obtain the log-likelihood function: The log-likelihood function is (3) The log-likelihood function is iterated using the gradient ascent method to obtain the maximum value of the objective function L(w). At this time, the logistic regression model determined by w is the prediction model for evaluating the postoperative survival rate of lung cancer patients.
2. The system according to claim 1, It is characterized in that The gradient direction of the gradient in step (3) is 3. The system according to claim 1, It is characterized in that The iteration in step (3) is: iterate w in the gradient direction, specifically:
4. The system according to claim 1, Features: The data collection module is used to obtain samples of lung cancer patients who undergo lung resection surgery; the lung cancer patient samples include at least 7 of the lung cancer patient's tumor type, forced vital capacity, volume of air exhaled in the first second of maximum exhalation, Zhu score, pre-operative pain, pre-operative hemoptysis, pre-operative dyspnea, pre-operative cough, pre-operative weakness, original tumor size, type 2 diabetes, peripheral arterial disease, smoking status, asthma status and patient age.
5. The system according to claim 1, Features: The data preprocessing module includes discrete feature encoding and standardization processing of the data.
6. The system according to claim 5, Features: The formula for the standardization process is: Among them, x represents the original eigenvalue, μ is the mean of the column, σ is the standard deviation of the column, and x* is the final eigenvalue.
7. The system according to claim 1, Features: The data of the data preprocessing module includes training set data, which is used to train a prediction model for evaluating the postoperative survival rate of lung cancer patients.
8. The system according to claim 7, Features: The training method is: substituting the training set data into the obtained lung cancer patient postoperative survival rate assessment prediction model, calculating the probabilities p(y=1|x) and p(y=0|x), and comparing the two probabilities, and taking the value of the item y with a larger probability as the category of the new sample.
9. The system according to claim 1, Features: The result output module is used to output the survival rate of lung cancer patients one year after lung resection.
10. The system according to any one of claims 1 to 9, Features: The method for predicting the survival rate of patients after lung cancer surgery by the system comprises the following steps: S1. Use the data collection module to obtain the sample data; S2, using a data preprocessing module to preprocess the data set selected in step S1; S3, inputting the data pre-processed in step S2 into the prediction and evaluation module; S4. Obtain an evaluation report on the survival rate of patients after lung cancer surgery through the result output module.
Citation Information
Patent Citations
Pancreas neuroendocrine lymph node transfer prediction system based on Logistic regression model
CN108492884A
