Vision Foundation Models in Radiology: A Scoping Review of Data, Methodology, Evaluation and Clinical Translation

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Alejandro Vergara-Richart, Xavier Rafael-Palou, Almudena Fuster-Matanzo, Ignacio Iborra Roncales, Ángel Alberich-Bayarri, Ana Jiménez-Pastor
• Venue: arXiv: cs.CV
• Published: 2026-07-08
• Collected: 2026-07-09
• URL: http://arxiv.org/abs/2607.07219v1
• PDF: https://arxiv.org/pdf/2607.07219v1
• DOI: UNKNOWN

Plain-Language Brief

The text discusses a scoping review of vision foundation models (VFMs) used in radiology. The study found that while these models are becoming common in radiological imaging, their development and evaluation methods are diverse. Most studies focused on brain MRI, thoracoabdominal CT, and chest X-ray datasets. Transformer-based architectures and self-supervised pretraining were prevalent, with masked image modeling, contrastive learning, and multi-stage approaches being popular. Evaluation primarily focused on segmentation and classification, but cross-validation was inconsistent. The study also noted that alignment with FUTURE-AI principles was uneven, and clinical translation is limited by data representativeness issues, heterogeneous benchmarks, incomplete reporting, and lack of deployment-oriented evaluation.

What the Document Says

• The study focused on vision foundation models (VFMs) for radiological imaging.
• Sixty-seven studies were included in the review.
• Datasets used primarily covered brain MRI, thoracoabdominal CT, and chest X-ray.
• Transformer-based architectures and self-supervised pretraining were common.
• Masked image modeling, contrastive learning, and multi-stage approaches were popular.
• Evaluation mainly focused on segmentation and classification.
• Cross-center, cross-scanner, anatomical, and modality-shift validation was inconsistently reported.
• Alignment with FUTURE-AI principles was uneven.
• Clinical translation is limited by data representativeness issues, heterogeneous benchmarks, incomplete reporting, and lack of deployment-oriented evaluation.

Key Concepts Mentioned

• Vision Foundation Models (VFMs)
• Radiology
• Data scale and heterogeneity
• Architectural and pretraining scalability
• Downstream transferability and generalization
• Transformer-based architectures
• Self-supervised pretraining
• Masked image modeling
• Contrastive learning
• Multi-stage approaches
• Segmentation
• Classification
• Cross-center validation
• Cross-scanner validation
• Anatomical validation
• Modality-shift validation
• FUTURE-AI principles

Claims Made by the Document

• VFMs show promising transferability.
• Clinical translation is constrained.

Evidence Used

• Sixty-seven studies were included and mapped across three pillars.
• Evaluation focused mainly on segmentation and classification.
• Cross-center, cross-scanner, anatomical and modality-shift validation was inconsistently reported.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Vision foundation models (VFMs) are increasingly being developed for radiological imaging, yet their definition, development and evaluation remain heterogeneous. We conducted a PRISMAScR scoping review of peer-reviewed studies published between January 2017 and March 2026 describing foundation models trained exclusively on radiological imaging data. Sixty-seven studies were included and mapped across three pillars: data scale and heterogeneity, architectural and pretraining scalability, and downstream transferability and generalization. Datasets primarily covered brain MRI, thoracoabdominal CT, and chest X-ray, ranging from fewer than 100,000 samples to multi-million-image cohorts. Transformer-based architectures and self-supervised pretraining predominated, particularly masked image modeling, contrastive learning and multi-stage approaches. Evaluation focused mainly on segmentation and classification, whereas cross-center, cross-scanner, anatomical and modality-shift validation was inconsistently reported. Alignment with FUTURE-AI principles was uneven. Overall, radiology-specific VFMs show promising transferability, but clinical translation remains constrained by limited data representativeness, heterogeneous benchmarks, incomplete reporting and insufficient deployment-oriented evaluation.