Method and system for predicting brain metastasis from non-small cell lung cancer, and apparatus
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- SHANGHAI PULMONARY HOSPITAL (SHANGHAI OCCUPATIONAL DISEASE PREVENTION & CONTROL INSTITUTE)
- Filing Date
- 2025-01-15
- Publication Date
- 2026-05-21
AI Technical Summary
Existing non-small cell lung cancer brain metastasis prediction models have poor sensitivity and specificity, making it difficult to effectively identify high-risk patients and affecting clinical treatment decisions and prognosis.
Patient data was obtained using the SEER database, and variables that were not statistically significant were removed using the LASSO algorithm. Various machine learning models were constructed, including logistic regression, random forest, and support vector machine. Gradient boosting machine models were used for training and validation, and visualization tools were developed for risk prediction.
It improves the accuracy and sensitivity of predicting the risk of brain metastasis in non-small cell lung cancer, helping clinicians to identify high-risk patients early, prolonging patients' life expectancy and improving their quality of life.
Smart Images
Figure CN2025072467_21052026_PF_FP_ABST
Abstract
Description
A method, system, and device for predicting brain metastases in non-small cell lung cancer. Technical Field
[0001] This invention relates to the field of cancer brain metastasis diagnosis technology, specifically to a method, system, and device for predicting brain metastasis in non-small cell lung cancer. Background Technology
[0002] Brain metastases (BM) are among the most serious neurological complications of cancer, contributing significantly to morbidity and mortality in affected patients. Common primary malignancies associated with brain metastases include lung cancer, breast cancer, and melanoma. In non-small cell lung cancer (NSCLC), approximately 10%–25% of patients present with brain metastases at initial diagnosis, and about 50% develop brain metastases during treatment. The reported incidence of brain metastases in NSCLC patients at initial diagnosis is 10% to 20%, but can be as high as 40% during clinical course. Patients with brain metastases generally have a poor prognosis, with a median survival of less than one year, and often experience poor quality of life and outcomes.
[0003] With advancements in treatment methods, many patients with non-small cell lung cancer (NSCLC) and brain metastases now benefit from stereotactic radiosurgery, a key component of a multimodal therapy designed to improve outcomes. However, some NSCLC brain metastases may be latent and asymptomatic. Furthermore, the prognosis for these patients with brain metastases is generally poor. Therefore, timely detection of brain metastases is essential for subsequent examination and treatment planning. Identifying risk factors for brain metastases is therefore crucial for improving predictive assessment and guiding clinical treatment.
[0004] In the field of cutting-edge artificial intelligence methods, machine learning (ML), including algorithms such as logistic regression (LR), classification and regression trees (CART), random forests (RF), support vector machines (SVM), k-nearest neighbors (KNN), gradient boosting machines (G-brain transfer), and extreme gradient boosting (XGBOOST), has become an innovative and widely used tool. Its importance lies in its ability to enhance the predictive accuracy of diagnostic processes and prognostic assessments. ML algorithms can autonomously learn knowledge from training datasets, promote dynamic interactions between variables, and identify potential key predictor variables. By recognizing patterns within the dataset, these algorithms establish optimal relationships between outcomes and potential predictor variables. This autonomous improvement in decision-making ability significantly enhances predictive effectiveness, demonstrating a significant improvement compared to other methods. Summary of the Invention
[0005] The purpose of this invention is to provide a method, system, and device for predicting brain metastasis in non-small cell lung cancer. It develops a prediction model through machine learning algorithms, assesses the risk probability of brain metastasis in non-small cell lung cancer, and establishes a convenient visualization tool to help clinicians predict the risk of brain metastasis in non-small cell lung cancer.
[0006] To achieve the above objectives, the present invention provides a method for predicting brain metastases in patients with non-small cell lung cancer, comprising:
[0007] Data acquisition: Patient data is acquired based on the SEER database, and the patient data includes a set of variables for patient baseline information;
[0008] Data preprocessing was performed to exclude variables with missing values, and the LASSO algorithm was used to remove patient data with statistically insignificant variables.
[0009] Model building involves randomly dividing the screened patient data into training, validation, and test sets, and constructing multiple machine learning models to achieve optimal predictive performance.
[0010] Statistical analysis was performed to train the machine learning model based on the training and validation sets. The training results were evaluated using the test set and based on the area under the ROC curve, calibration curve, and model performance parameters. The model with the best performance was then selected.
[0011] Furthermore, the baseline information of the patient data includes age, gender, marital status, histological type, tumor size, tumor laterality, surgical history, radiotherapy history, chemotherapy history, T stage, N stage, distant lymph node metastasis, brain metastasis, and lung metastasis.
[0012] Furthermore, based on the LASSO algorithm, the baseline information of the patient data includes: age, marital status, histological type, tumor size, surgical history, radiotherapy history, chemotherapy history, T stage, N stage, distant lymph node metastasis, brain metastasis, and lung metastasis.
[0013] Furthermore, the machine learning models include logistic regression models, classification and regression trees, random forests, support vector machines, K-nearest neighbors, gradient boosting machines, and extreme gradient boosting machines.
[0014] Furthermore, based on statistical analysis, the AUC values of the gradient booster model on the validation set and the test set are 0.8276 and 0.8301, respectively; the model performance parameters are: accuracy: 0.8752, sensitivity: 0.890, specificity: 0.5595, Kappa: 0.224; accuracy: 0.8768, sensitivity: 0.8991, specificity: 0.5826, Kappa: 0.2280. The gradient booster model is the best performing model.
[0015] Furthermore, based on existing models, visualization tools can be developed to predict the risk of brain metastases by inputting patients' clinical information.
[0016] On the other hand, the present invention also provides a brain metastasis prediction system for patients with non-small cell lung cancer, comprising:
[0017] The data acquisition module acquires patient data based on the SEER database, and the patient data includes a set of variables of patient baseline information;
[0018] The data preprocessing module excludes variables with missing values and uses the LASSO algorithm to remove patient data with statistically insignificant variables.
[0019] The model building module randomly divides the screened patient data into training, validation and test sets, and builds multiple machine learning models to achieve the best predictive performance.
[0020] The statistical analysis module trains machine learning models based on training and validation sets, evaluates the training results using a test set and based on the area under the ROC curve, calibration curve, and model performance parameters, and selects the model with the best performance.
[0021] On the other hand, the present invention also provides a storage device, the storage medium storing a plurality of instructions adapted for loading by a processor to execute the steps in the above-described method for predicting brain metastases in patients with non-small cell lung cancer.
[0022] This invention provides a method, system, and device for predicting brain metastasis in non-small cell lung cancer (NSCLC). Specifically designed for assessing the risk of brain metastasis (BM) in NSCLC patients, it utilizes lung cancer patient data from the SEER database and employs various machine learning algorithms to construct a model predicting the risk of brain metastasis in NSCLC patients. Testing shows that the gradient booster model outperforms other classification models on both the test and validation sets. This model also achieves the highest accuracy, sensitivity, and Kappa value, while having the lowest Brier score. Furthermore, this invention provides visualization tools that can be used clinically to assist physicians in decision-making, early identification of high-risk patients, and advance planning of treatment and care. Attached Figure Description
[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0024] Figure 1 is a flowchart of a method for predicting brain metastases in non-small cell lung cancer according to an embodiment of the present invention.
[0025] Figure 2 is a system framework diagram of a non-small cell lung cancer brain metastasis prediction system according to an embodiment of the present invention.
[0026] Figure 3 is a Lasso regression analysis diagram of an embodiment of the present invention.
[0027] Figure 4 shows the ROC curves of various machine learning models according to an embodiment of the present invention. Detailed Implementation
[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0029] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.
[0030] If the application documents contain similar descriptions such as "first / second", the following explanation shall be added: In the following description, the terms "first / second / third" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.
[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.
[0032] The poor prognosis of lung cancer with brain metastases has been a major concern. A large body of literature has explored prognostic factors for brain metastases (BM), including age, sex, histological grade, treatment, marital status, distant metastases, and tumor size. However, most studies have focused on survival assessment in lung cancer patients with BM. Assessing risk factors for BM in non-small cell lung cancer (NSCLC) patients is crucial for identifying high-risk patients, which is essential for considering candidates for prophylactic cranial irradiation (PCI) or other prophylactic therapies. Previous studies have identified several risk factors for BM in NSCLC, including younger age, non-squamous cell carcinoma, and disease stage. Furthermore, EGFR and ALK gene mutations and advanced disease have been identified as independent factors for BM development. High N stage is also considered a risk factor for BM, and lymph node invasion is associated with an increased risk of BM. Further research has shown that high serum CEA levels, NSE, and CA125 are also independent risk factors for BM. In small cell lung cancer patients, higher T stage, higher NLR, earlier thoracic radiotherapy, and fewer chemotherapy cycles are associated with an increased risk of BM. Consistent with the above findings, our study also demonstrates the important role of tumor size and bone metastasis in bone marrow metastasis (BM) in NSCLC patients.
[0033] In the prior art, developing predictive models for distant organ metastasis in lung cancer patients has the potential to identify individuals with clinical characteristics more prone to metastasis. Currently, many predictive models for metastasis (BM) in non-small cell lung cancer (NSCLC) patients focus primarily on clinical factors. However, previously reported models have limitations in sensitivity and specificity, posing a significant challenge to predictive capabilities. These findings suggest that clinicians should consider clinicopathological features when developing diagnostic and treatment algorithms. This invention not only helps identify high-risk patients but also facilitates the development of individualized assessments and personalized examination measures. Ultimately, this invention can extend patients' life expectancy, improve their quality of life, and reduce the economic burden on society and families.
[0034] The following describes, with reference to the accompanying drawings, a method, system, and storage device for assessing the cardiotoxicity of immune checkpoint inhibitors according to an embodiment of the present invention. First, the method for assessing the cardiotoxicity of immune checkpoint inhibitors according to an embodiment of the present invention will be described with reference to the accompanying drawings.
[0035] Figure 1 is a flowchart of a method for predicting brain metastases in non-small cell lung cancer according to an embodiment of the present invention. As shown in Figure 1, the prediction method includes the following steps:
[0036] S1 data acquisition involves acquiring patient data based on the SEER database, which includes a set of variables representing patient baseline information.
[0037] S2 data preprocessing: variables with missing values were excluded, and the LASSO algorithm was used to remove patient data with statistically insignificant variables.
[0038] The S3 model is constructed by randomly dividing the screened patient data into training, validation, and test sets, and building multiple machine learning models to achieve the best predictive performance.
[0039] S4 statistical analysis is used to train machine learning models based on training and validation sets, and to evaluate the training results using a test set based on the area under the ROC curve, calibration curve, and model performance parameters, selecting the model with the best performance.
[0040] Specifically, in step S1, patient data is obtained from the Surveillance, Epidemiology, and End Outcomes (SEER) project (www.seer.cancer.gov).
[0041] In one specific embodiment, the inclusion criteria for patient data included: (1) patients diagnosed between January 2010 and December 2020; (2) the location and morphology of the lungs and bronchi according to ICD-O-3 / WHO-2008 classification; (3) ICD-O-3 behavioral codes indicating malignancy; (4) histological type of non-small cell lung cancer determined according to the International Classification of Diseases oncology (3rd edition) code; and (5) the presence of a malignant primary lesion.
[0042] Exclusion criteria for patient data included: (1) T stage or in situ tumor (Tis) or undetected (T0) lacked specificity; (2) N stage lacked specificity; and (3) metastasis lacked specificity.
[0043] In one embodiment, the clinicopathological records of NSCLC patients were extracted using SEER*stat software (version 8.4.2). The final extracted baseline patient data included age, sex, marital status, histological type, tumor size, tumor laterality, surgical history, radiotherapy history, chemotherapy history, T stage, N stage, distant lymph node metastasis, brain metastasis, and lung metastasis.
[0044] Specifically, in step S2, after excluding variables with missing values in the patient data, the Least Absolute Contraction and Selection Operator (LASSO) regression analysis is used to remove variables that are not statistically significant.
[0045] Figure 3 is a Lasso regression analysis diagram according to an embodiment of the present invention, wherein the left figure shows the evaluation of the variability of variable coefficients, and the right figure shows the process of determining the optimal λ parameter in the Lasso regression model using the cross-validation method.
[0046] After being filtered by the LASSO algorithm, the baseline information of the patient data includes: age, marital status, histological type, tumor size, surgical history, radiotherapy history, chemotherapy history, T stage, N stage, distant lymph node metastasis, brain metastasis, and lung metastasis.
[0047] Specifically, in step S3, this embodiment randomly divides 39,930 patients into three parts: a training set (60% of participants), a validation set (20% of participants), and a test set (20% of participants). Based on the above data, this embodiment constructs seven machine learning models, including Logistic Regression (LR), Classification and Regression Tree (CART), Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Gradient Boosting Machine (GBM), and Extreme Gradient Boosting (XGBOOST), to achieve optimal predictive performance.
[0048] In step S4, continuous variables in this embodiment are represented by the median, and categorical variables are represented by numbers and percentages. The accuracy of the training cohort model is tested on the test and validation cohort data. The area under the ROC curve is used to distinguish the model's ability across the two datasets. All statistical analyses, model building, and validation in this study were performed using R software (version 4.1.3).
[0049] The predictive performance of the model is evaluated using the receiver operating characteristic (ROC) curve, with a higher area under the curve (AUC) value indicating stronger discriminative ability.
[0050] Figure 4 shows the ROC curves of various machine learning models according to an embodiment of the present invention. The left figure shows the ROC curves for the validation queue, and the right figure shows the ROC curves for the test queue. As shown, the AUC values in the validation set are 0.7818, 0.7286, 0.7824, 0.7250, 0.7341, 0.8276, and 0.8227, respectively. In the test set, the AUC values are 0.7907, 0.7165, 0.7793, 0.7425, 0.7333, 0.8301, and 0.8295, respectively.
[0051] In this embodiment, the model performance parameters of each machine learning model are shown in the table below:
[0052] Based on the above analysis and statistical analysis, the AUC values of the gradient booster model on the validation set and the test set are 0.8276 and 0.8301, respectively; the model performance parameters are: accuracy: 0.8752, sensitivity: 0.890, specificity: 0.5595, Kappa: 0.224; accuracy: 0.8768, sensitivity: 0.8991, specificity: 0.5826, Kappa: 0.2280. The gradient booster model is the best performing model.
[0053] In one specific embodiment, to facilitate use by medical personnel during work, the present invention employs a mobile device or computer as an electronic device for running the method of the present invention. This electronic device includes a processor and a memory for storing processor-executable instructions. The executable instructions stored in the memory are configured to run a visualization program for predicting brain metastases in non-small cell lung cancer according to the present invention. This visualization program can communicate with other terminals, servers, or other types of devices to complete background program calculations.
[0054] Using visualization tools, clinicians can input clinical information of a typical patient into the model, such as a 75-year-old single male with adenocarcinoma in the main bronchus, a tumor size of 61 mm, T4N3, no surgery or chemotherapy, a history of radiotherapy, no lymph node metastasis but lung metastasis. The model predicts this patient's risk of BM (bulbar bronchitis) at 57.48%, indicating a high risk of BM.
[0055] Figure 2 is a system framework diagram of a non-small cell lung cancer brain metastasis prediction system according to an embodiment of the present invention. As shown in Figure 2, a brain metastasis prediction system for non-small cell lung cancer patients includes:
[0056] Data acquisition module 1 acquires patient data based on the SEER database, the patient data including a set of variables of patient baseline information;
[0057] Data preprocessing module 2 excludes variables with missing values and uses the LASSO algorithm to remove patient data with statistically insignificant variables;
[0058] Model building module 3 randomly divides the screened patient data into training set, validation set and test set, and builds multiple machine learning models to achieve the best prediction performance.
[0059] Statistical analysis module 4 trains machine learning models based on training and validation sets, evaluates the training results using a test set and based on the area under the ROC curve, calibration curve, and model performance parameters, and selects the model with the best performance.
[0060] Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for predicting brain metastases in patients with non-small cell lung cancer, characterized in that, include: Data acquisition: Patient data is acquired based on the SEER database, and the patient data includes a set of variables for patient baseline information; Data preprocessing was performed to exclude variables with missing values, and the LASSO algorithm was used to remove patient data with statistically insignificant variables. Model building involves randomly dividing the screened patient data into training, validation, and test sets, and constructing multiple machine learning models to achieve optimal predictive performance. Statistical analysis was performed to train the machine learning model based on the training and validation sets. The training results were evaluated using the test set and based on the area under the ROC curve, calibration curve, and model performance parameters. The model with the best performance was then selected.
2. The method for predicting brain metastases in non-small cell lung cancer patients as described in claim 1, characterized in that, The baseline information of the patient data includes age, sex, marital status, histological type, tumor size, tumor laterality, surgical history, radiotherapy history, chemotherapy history, T stage, N stage, distant lymph node metastasis, brain metastasis, and lung metastasis.
3. The method for predicting brain metastases in non-small cell lung cancer patients as described in claim 2, characterized in that, Based on the LASSO algorithm, the baseline information of the patient data includes: age, marital status, histological type, tumor size, surgical history, radiotherapy history, chemotherapy history, T stage, N stage, distant lymph node metastasis, brain metastasis, and lung metastasis.
4. The method for predicting brain metastases in non-small cell lung cancer patients as described in claim 3, characterized in that, The machine learning models include logistic regression, classification and regression trees, random forests, support vector machines, K-nearest neighbors, gradient boosting machines, and extreme gradient boosting machines.
5. The method for predicting brain metastases in non-small cell lung cancer patients as described in claim 4, characterized in that, Based on statistical analysis, the AUC values of the gradient booster model on the validation set and the test set are 0.8276 and 0.8301, respectively; the model performance parameters are accuracy and... 0.8752, Sensitivity: 0.890, Specificity: 0.5595, Kappa: 0.224; Accuracy: 0.8768, Sensitivity: 0.8991, Specificity: 0.5826, Kappa: 0.2280, Gradient booster model is the best performing model.
6. The method for predicting brain metastases in non-small cell lung cancer patients as described in claim 5, characterized in that, A visualization tool was developed based on existing models to predict the risk of brain metastasis in patients by inputting their clinical information.
7. A brain metastasis prediction system for non-small cell lung cancer patients, characterized in that, include: The data acquisition module acquires patient data based on the SEER database, and the patient data includes a set of variables of patient baseline information; The data preprocessing module excludes variables with missing values and uses the LASSO algorithm to remove patient data with statistically insignificant variables. The model building module randomly divides the screened patient data into training, validation and test sets, and builds multiple machine learning models to achieve the best predictive performance. The statistical analysis module trains machine learning models based on training and validation sets, evaluates the training results using a test set and based on the area under the ROC curve, calibration curve, and model performance parameters, and selects the model with the best performance.
8. A storage device, characterized in that, The storage medium stores a plurality of instructions adapted for loading by a processor to perform steps in the method for evaluating the cardiotoxicity of immune checkpoint inhibitors according to any one of claims 1 to 6.