Reproduce “Vision Foundation Models in Radiology: A Scoping Review of Data, Methodology, Evaluation and Clinical Translation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Alejandro Vergara-Richart, Xavier Rafael-Palou, Almudena Fuster-Matanzo, Ignacio Iborra Roncales, Ángel Alberich-Bayarri, Ana Jiménez-Pastor
A scoping review by Vergara-Richart et al. from Quibim S.L. systematically maps the landscape of Vision Foundation Models (VFMs) in radiology, detailing their diverse data characteristics, predominant methodological approaches like transformer architectures and self-supervised learning, and varied evaluation strategies, while also assessing alignment with clinical trustworthiness principles.
Key insights
- Radiology VFM development is characterized by rapid growth and significant methodological heterogeneity, including highly variable data scales and compositions (e.g., 58.2% single modality).
- Transformer architectures are predominant (79.1%), and self-supervised learning, particularly masked image modeling (40%), is the leading pretraining strategy, with models typically having a median of 214 million parameters.
- While most studies evaluate multiple downstream tasks (median 6) and various generalization shifts, assessment of trustworthiness principles like fairness (50.7%) and explainability (71.6%) remains inconsistent compared to robustness (94%).
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.