Multi-View Radiomic Feature Analysis for Machine Learning-Based Lung Cancer Classification

Radiological analysis is important for getting clear, quantitative information from medical images. It helps us understand differences within tumor tissues, including their shapes and densities. In this study, we first assessed statistical, morphological, and textural radiomic features individually, considering each as a separate viewpoint for analysis. These three independent views were then combined using a multi-view approach: early fusion, late fusion, and intermediate fusion. At each classification stage, analysis was performed using Random Forest Classifier, Decision Tree Classifier, K-Nearest Neighbors, Logistic Regression, Support Vector Machine, Hist Gradient Boosting Classifier, and Multilayer Perceptron Classifier algorithms. The study results show that statistical and textural features are effective for discrimination, whereas morphological features are less effective when used alone. Multiple-view learning strategies combining different radiomic appearances yield significantly better classification performance than single-view approaches. This study highlights the need for a holistic evaluation of heterogeneous feature sets to determine radiomics-based cancer subtypes. It demonstrates that successful results can be achieved in multi-view radiomics analyses.