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

This PRISMA-ScR scoping review included 67 peer-reviewed studies published from January 2017 through March 2026 on foundation models trained exclusively with radiological images. The literature was mapped across data scale and heterogeneity, architectural and pretraining scalability, and downstream transferability and generalization. Brain MRI, thoracoabdominal CT, and chest X-ray dominated the datasets, while Transformers, self-supervised learning, masked image modeling, contrastive learning, and multi-stage pretraining were common. Evaluation centered on segmentation and classification, but cross-center, cross-scanner, anatomical, and modality-shift validation was inconsistently reported. The review concludes that clinical translation remains limited by representativeness, heterogeneous benchmarks, incomplete reporting, and insufficient deployment-oriented evaluation.