Nvidia releases research model for three-dimensional CT analysis

Nvidia has made checkpoints available for NV-Reason-CT, an open research model built to analyze three-dimensional CT scans and generate structured reports. Developers can use the checkpoints and published scripts to test or adapt the model for chest and abdominal imaging. A three-dimensional image encoder preserves spatial relationships between scan slices, while a language model produces reports and answers follow-up questions. Nvidia says the model outperforms published baselines on the CT-RATE benchmark, but that performance claim comes from the company. NV-Reason-CT is a research foundation for further development, not an autonomous diagnostic system or a cleared clinical product.