Foundation Models vs. Radiomics for Lung Computed Tomography: A Benchmark of Feature Extractors, Classification Heads, and Segmentation Choices

2026-07-01

AI Summary

This study addresses the lack of systematic, disentangled comparisons between foundation models and radiomics in lung CT analysis, particularly regarding cross-cohort robustness. Through a two-stage design, the authors evaluate diverse combinations—including feature extractors (Curia, DINOv3, Radiomics), classifiers (TabPFN, XGBoost, CatBoost), and segmentation strategies (tumor vs. whole-lung)—across five clinical tasks, using worst-case cross-cohort performance as the primary metric. The work presents the first disentangled analysis of individual component contributions, revealing that segmentation predominantly influences tumor volume and staging tasks, while the classifier dominates survival, histology, and age prediction. Task-dependent design principles are proposed, with Curia combined with tumor segmentation and CatBoost recommended as a default configuration, demonstrating superior average performance across three core tasks.

Abstract

Radiomics is the established approach for CT-based lung cancer phenotyping, yet comparisons with foundation models rarely isolate contributions of feature extractor, classification head, and segmentation choice, or test cross-cohort robustness. We benchmark five feature extractors (Curia, Curia-2, DINOv3, Radiomics2D, Radiomics3D), seven classification heads (TabPFN, TabICL, XGBoost, CatBoost, Random Forest, logistic regression, Ridge), and three segmentation regimes on five tasks: tumor volume and stage classification, 2-year survival prediction, histology classification, and age prediction. Models are trained on LUNG1 (n=338) and evaluated on an internal test set (n=84) and the external LUNG2 cohort (n=211), with worst-case cross-cohort performance as the primary metric. The dominant design factor is task-dependent: segmentation drives volume and stage classification, while classifier choice drives survival, histology, and age prediction. Radiomics is competitive for tumor volume, tumor stage and survival (partly due to label-derivation effects for the former); Curia variants reach comparable peak scores for survival; DINOv3 falls slightly short across tasks. Patch and slice aggregation have negligible impact. We recommend Curia with tumor segmentation and a CatBoost head as a safe default, achieving the best mean rank across the three primary clinical tasks, though task-specific selection consistently outperforms any cross-task default. When tumor delineations are unavailable, Curia-2 with lung segmentation and logistic regression offers a competitive alternative. All pipelines use a two-stage design suited to small cohort sizes where end-to-end fine-tuning would risk overfitting.

Problem

• Foundation Models
• Radiomics
• Lung CT
• Cross-cohort Robustness
• Feature Extractors

Innovation

• foundation models
• radiomics
• cross-cohort robustness
• two-stage design
• feature extractor benchmarking