NVIDIA launches NV-Reason-CT Open 3D CT VLM for radiologist chain-of-thought reasoning
The model is designed to improve systematic reasoning in 3D CT scans. It builds on prior research in explainable AI for radiology. The system is still in development and will be available in 2026.
NVIDIA has introduced NV-Reason-CT Open, a 3D CT vision language model (VLM) designed to support radiologist chain-of-thought reasoning. The system aims to address the limitations of existing AI models in handling the complexity of volumetric imaging. By providing a step-by-step reasoning process, NV-Reason-CT reflects how radiologists think through a CT study, offering a more systematic and interpretable approach to medical imaging analysis.
Radiology AI has made significant progress in detecting abnormalities in chest X-rays, pathology slides, and 2D scans. However, 3D computed tomography (CT) scans remain underserved by current vision language models. These scans are data-dense and clinically rich, requiring a different approach to analysis that standard 2D encoders cannot fully capture. NVIDIA is addressing this gap with NV-Reason-CT, a VLM purpose-built for 3D CT analysis.
NV-Reason-CT is built on the reasoning methodology pioneered by NV-Reason-CXR, which was validated in multiple real-world applications. The model uses a 3D ViT architecture and incorporates a 3D MRoPE mechanism to handle the spatial complexity of volumetric imaging. This approach allows the model to process up to 192³ voxels, enabling more accurate and context-aware reasoning across anatomical structures.
The introduction of NV-Reason-CT could lead to higher costs for healthcare institutions adopting the technology, as well as potential vendor lock-in with NVIDIA's ecosystem. Governance and regulatory considerations will also play a role in its deployment, as medical AI systems require rigorous validation and compliance with industry standards. Market reactions may vary, with some institutions embracing the model's capabilities while others may prefer to wait for further development and testing.
NV-Reason-CT is still in development and is expected to be available in 2026. The model is part of NVIDIA's broader effort to advance explainable AI in radiology, building on previous research such as NV-Segment-CTMR and NV-Reason-CXR. As the system evolves, it may influence the future of AI-driven diagnostic tools in medical imaging, potentially reshaping how clinicians interact with and rely on AI for complex 3D CT analysis.