Iconic Research and Engineering Journals
Patient-Level Deep Fusion, Radiomics, and Ensemble Learning for CT-Based Differentiation of Multiple Primary Lung Cancers
By Arnab Dutta.
Abstract
Background
Multiple primary lung cancers (MPLCs) are increasingly detected owing to advances in chest computed tomography (CT), yet distinguishing them from intrapulmonary metastases (IPMs) and multiple benign pulmonary lesions (MBPLs) remains difficult and strongly affects treatment planning.
Methods
We propose a patient-level framework that combines (i) a deep fusion network integrating 2D multiple-instance learning (2D-MIL) over axial slices, a 3D volumetric branch, and handcrafted radiomic features, with (ii) radiomics-based Random Forest and XGBoost classifiers, and (iii) a soft-voting ensemble of these models. Performance was assessed with accuracy, precision, recall, F1-score, specificity, and the area under the receiver operating characteristic curve (ROC-AUC).
