Breaking

NVIDIA Unveils NV-Reason-CT: AI Model Generates Explainable Radiology Reports from 3D CT Scans

October 12, 2026 at 01:15 EDT

Healthcare
Foundation Models
Open Source

NVIDIA has released NV-Reason-CT, a 4.69-billion-parameter vision-language model that generates structured diagnostic reports and radiologist-style reasoning from three-dimensional computed tomography volumes. The open-source foundation model, announced on NVIDIA's Developer Blog, extends chain-of-thought reasoning capabilities to volumetric medical imaging, enabling AI systems to articulate diagnostic logic step-by-step for clinical transparency and verification.

Medical AI · October 2026

NVIDIA Releases Explainable 3D CT Reasoning Model

NV-Reason-CT generates diagnostic reports with step-by-step radiologist-style reasoning from volumetric CT scans, improving both transparency and clinician efficiency.

4.69B

Model Parameters

550K+

Training QA Examples

0.614

CT-RATE Macro-F1

50%

Faster Reporting

Architecture Approach

→

3D Vision Transformer

Processes 192³ mm volumes natively, generating 13,824 vision tokens

→

3D Positional Embedding

3D MRoPE preserves spatial relationships through language model

→

Qwen 3.5-4B Language Model

Generates structured reports with chain-of-thought reasoning

✓ Strengths

• True 3D processing preserves anatomy
• Transparent reasoning enhances trust
• 50% faster reporting in studies
• Open source on GitHub & Hugging Face

⚠ Limitations

• Can hallucinate unsupported findings
• Fixed anatomical crop selection
• Ignores clinical context & priors
• Preliminary clinical validation

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.