Radiomics-Based Lung Nodule Classification with Stacking Ensembles

Radiomics, an emerging field in medical imaging, leverages advanced mathematical analysis to extract quantitative metrics from medical images, aiding in the early detection, diagnosis, and treatment of lung cancer. This study focuses on improving the risk prediction of small lung nodules using machine learning models. We employed a stacking ensemble approach, integrating Principal Component Analysis (PCA) for dimensionality reduction and selected features from the Small Nodule Radiomics-Predictive Vector (SN-RPV). Base models employed in the stacking ensemble were Support Vector Machine (SVM), Random Forest, k-Nearest Neighbors (KNN), and Naive Bayes classifiers. Despite the theoretical advantages of stacking ensembles, our models demonstrated poorer performance on the test set compared to the simpler SN-RPV model by Hunter et al. This outcome highlights the challenges of overfitting and underscores the importance of model simplicity and interpretability in clinical applications. Future research should explore alternative regularization techniques to improve the generalization of complex ensemble methods.