Curia-2: Scaling Self-Supervised Learning for Radiology Foundation Models
This work addresses the unsustainable workload on radiologists by developing more efficient foundation models for medical imaging, representing an incremental advancement in domain-specific AI.
The paper tackles the problem of optimizing foundation models for radiology by introducing Curia-2, which improves pre-training strategies and scales to billion-parameter vision transformers, outperforming other foundation models on vision tasks and competing with vision-language models on complex clinical tasks like finding detection.
The rapid growth of medical imaging has fueled the development of Foundation Models (FMs) to reduce the growing, unsustainable workload on radiologists. While recent FMs have shown the power of large-scale pre-training to CT and MRI analysis, there remains significant room to optimize how these models learn from complex radiological volumes. Building upon the Curia framework, this work introduces Curia-2, which significantly improves the original pre-training strategy and representation quality to better capture the specificities of radiological data. The proposed methodology enables scaling the architecture up to billion-parameter Vision Transformers, marking a first for multi-modal CT and MRI FMs. Furthermore, we formalize the evaluation of these models by extending and restructuring CuriaBench into two distinct tracks: a 2D track tailored for slice-based vision models and a 3D track for volumetric benchmarking. Our results demonstrate that Curia-2 outperforms all FMs on vision-focused tasks and fairs competitively to vision-language models on clinically complex tasks such as finding detection. Weights will be made publicly available to foster further research.