Foundation Models Benchmarked Against Radiomics for Lung CT Analysis

A new benchmark study published on arXiv compares foundation models against traditional radiomics techniques for analyzing lung CT scans. The research evaluated five feature extractors, seven classification heads, and three segmentation approaches across five tasks, including tumor classification and survival prediction. Findings indicate that segmentation is crucial for volume and stage classification, while the choice of classifier significantly impacts survival and histology prediction. The study suggests a default pipeline using Curia with tumor segmentation and CatBoost for clinical tasks, while also offering an alternative when tumor delineations are absent. AI

IMPACT This research provides a benchmark for AI models in medical imaging, guiding the selection of feature extractors and classifiers for lung CT analysis.

RANK_REASON The item is a research paper published on arXiv detailing a benchmark study comparing AI models and traditional methods. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• CatBoost
• DINOv3
• logistic regression
• LUNG1
• LUNG2
• Ridge
• Radiomics2D
• Radiomics3D
• Random Forest
• TabICL
• TabPFN
• XGBoost

• paper
• other

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →