Curia: A Multi-Modal Foundation Model for Radiology (arxiv.org)

Curia is a newly introduced multi-modal foundation model specifically designed for radiology, addressing the limitations of current AI systems that typically focus on narrow, single-task applications. Trained on an unprecedented dataset of 150,000 cross-sectional imaging exams (130 TB) from a major hospital, Curia demonstrates broad capabilities across diverse imaging modalities and diagnostic tasks. Its large-scale, real-world training corpus enables it to generalize effectively even in low-data scenarios, a critical challenge in medical AI. On a newly curated 19-task external validation benchmark, Curia excels at identifying organs, detecting critical conditions like brain hemorrhages and myocardial infarctions, and predicting tumor staging outcomes. Impressively, it matches or outperforms radiologists and recent foundation models, exhibiting emergent clinical properties across different imaging modalities. By releasing the base model’s weights publicly, the authors aim to foster further research and innovation in AI-driven radiological interpretation, potentially accelerating adoption of versatile, high-performing AI tools in medical imaging.