Foundation models in radiology: a primer for pediatric radiologists

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Pediatric Radiology

Background:

• Foundation models (FMs) are large deep learning models pre-trained on vast datasets.
• In pediatric radiology, data scarcity and anatomical variability pose challenges for traditional AI.
• FMs offer a robust mechanism for broad feature learning adaptable to specific clinical needs.

Purpose of the Study:

• To review the principles, architectures, and applications of FMs in pediatric radiology.
• To identify current and emerging applications of FMs in pediatric imaging.
• To outline challenges and future directions for FM integration in clinical practice.

Main Methods:

• Review of foundation model principles and architectures.
• Analysis of current and emerging applications in pediatric radiology.
• Identification of challenges and future perspectives for FM implementation.

Main Results:

• FMs can facilitate tasks like pathology detection, segmentation, and report generation in pediatric imaging.
• Potential benefits include improved diagnostic accuracy, workflow efficiency, and decision support.
• Key challenges include pediatric-specific disease spectra, limited datasets, ethical issues, and lack of validation.

Conclusions:

• FMs show promise for transforming pediatric imaging and advancing child-centered healthcare.
• Addressing challenges like data limitations, ethical concerns, and model explainability is crucial.
• Future techniques like federated learning and synthetic data generation may overcome implementation barriers.