August 12, 2025Open Access

Nira-03 Prediction of Lung Adenocarcinoma Brain Metastases Utilizing a Ct-Based Radiomic Model

Puntos clave

The predictive model achieved 79% accuracy for identifying brain metastases in lung adenocarcinoma patients, improving early detection.
Data analysis included 480 patients with treatment-naïve CT scans, resulting in 241 patients with brain metastases.
Feature selection used methods like RFE and classifiers such as Random Forest, creating robust predictive models.
Validation in external cohorts could support future clinical trials for early-stage lung adenocarcinoma patients needing MRI surveillance.

Resumen

Abstract Although almost 40% of NSCLC patients will develop BM during the course of their disease, there are currently no reliable prediction tools for identifying patients at risk for BM, especially in the early-stage setting where MRI screening is not performed. Furthermore, in later-stage cases, brain MRIs are only conducted annually. Identifying high-risk patients for BM who could benefit from MRI surveillance is crucial to enable earlier detection and intervention, potentially improving patient outcomes. We identified 480 lung adenocarcinoma (LUAD) patients with (N=241) or without (N=239) BM who had treatment-naïve CT scans with a segmentable lesion. The tumor, surrounding ground glass opacity, and necrosis were segmented via 3D Slicer to create a volume of interest for radiomic texture analysis, extracting 520 features. Training and testing sets were split 80:20, with a 5-fold cross-validation applied to the training set. We evaluated multiple feature selection methods (RFE, LASSO, ElasticNet, SelectKBest, and SelectFromModelRF) and classifier algorithms (Random Forest, XGBoost, SVM, Logistic Regression, Naive Bayes, and Neural Networks) to build our predictive models. The final models were selected based on the highest predictive performance. For predicting BM, the optimal model utilized RFE for feature selection and Random Forest as the classifier, yielding 79% accuracy, 77% sensitivity, 82% specificity, and 78% AUC in the overall population. For predicting metachronous BM only, the best model combined the SelectFromModelRF feature selection method with an XGBoost classifier, achieving 78% accuracy, 79% sensitivity, 76% specificity, and 78% AUC (p0.05). Data on the predictive value of this model across stage, molecular subtype, and extracranial metastases, as well as its correlation with overall survival, will be presented. Successful validation of this model in an external cohort will provide the rationale for a future trial of MRI surveillance in early-stage NSCLC patients.

Preguntar a la IA

Me gusta

Guardar

Ver artículo completo