{"resourceId":"hf-nv-reason-ct-20260923","versions":[{"version":"external-775c940a26ece947e676b4f1f6ffeb058427c499c3bcc07db14cc37b7a404e79","resource":{"id":"hf-nv-reason-ct-20260923","title":"NV-Reason-CT: open 3D vision-language model for chest/abdominal CT reasoning","organization":"NVIDIA","sector":"Medical imaging AI / open model release","geography":"Global (open weights; clinical transfer heavily constrained)","publishedAt":"2026-09-23T00:00:00Z","publicationDate":"2026-09-23","eventDate":"2026-09-23","sourceName":"Hugging Face model card","sourceLabel":"Vendor open model card","sourceUrl":"https://huggingface.co/nvidia/NV-Reason-CT","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["knowledge-work","developers-agents","governance-procurement"],"finding":"NVIDIA’s NV-Reason-CT Hugging Face card (release metadata around Sep 23, 2026; openmdw-1.1) describes a generative 3D vision-language model for chest and abdominal CT that pairs a Primus 3D ViT with a Qwen3.5-4B language model and is designed for radiology report generation, general question answering, and multi-step reasoning. Companion arXiv:2609.27511 was submitted 23 Sep 2026. This is a research/open-weights release, not an FDA-cleared clinical product claim on the card text reviewed.","sledRelevance":"Health-science and university medical centers may evaluate research use under IRB/clinical governance; do not treat as production diagnostic software from this card alone.","evidence":"HF card architecture and task statements; arXiv companion submitted 23 Sep 2026 inspected separately for date corroboration.","architectureImplications":"Native 3D CT encoders with large visual-token contexts differ from 2D slice VLM pipelines; need DICOM preprocessing and GPU memory planning.","governanceImplications":"Clinical research use requires IRB, PHI controls, and clear non-diagnostic labeling in procurement language.","securityPrivacyImplications":"CT volumes are PHI; inference logs and model outputs need HIPAA-aligned controls where applicable.","caveats":"Vendor model card; clinical performance claims require reading full paper/evals; not evidence of regulatory clearance.","streamIds":["nvidia"],"roles":{"sales":"Frame NV-Reason-CT as a research open-weights CT VLM—not a radiology product sale. Open with academic medical-center research imaging needs, then cite the card’s Primus 3D ViT plus Qwen3.5-4B design for report generation, QA, and multi-step reasoning. Bound any engagement to IRB-governed research sandboxes with PHI controls; explicitly list regulatory clearance and diagnostic accuracy as unsupported from this card alone. Keep non-diagnostic labeling in every slide and SOW from day one.","engineering":"Fit is research clusters with DICOM pipelines and GPU memory for 3D visual tokens; weak for production PACS integration without clinical engineering. Integrate offline research inference behind PHI-safe storage; pin openmdw-1.1 license obligations. Prerequisites: de-identification or IRB dataset, GPU nodes, DICOM converters. Constraints: clinical validation not established here; large 3D contexts. Proof: research-only report-quality eval on approved datasets with radiologist review protocol.","delivery":"Work: legal/license review, IRB pathway, PHI architecture, research deployment, evaluation protocol with clinicians. Dependencies: compliance, imaging IT, GPU capacity. Ownership: clinical research informatics. Skills: medical imaging ML ops and HIPAA operations. Governance: IRB and security review before any PHI. Acceptance: sandboxed research demo on approved data with non-diagnostic labeling. Risks: accidental clinical use, PHI leakage, and license misunderstandings by end users."},"roleDetails":{"sales":{"customerProblem":"Research groups want open 3D CT reasoning models without buying opaque clinical suites first.","stakeholders":["radiology research","clinical informatics","compliance/IRB","research computing"],"discoveryQuestions":["Is the use research-only or clinical care?","What PHI pathway exists?","Who signs non-diagnostic disclaimers?"],"valueHypothesis":"Open NV-Reason-CT can accelerate research prototyping if governance walls hold.","potentialEngagement":"IRB-scoped research sandbox on de-identified CT with radiologist-in-the-loop eval.","unsupportedClaims":["FDA/clinical clearance","Guaranteed report accuracy","Safe unattended clinical deployment"]},"engineering":{"fit":"Research 3D VLM stacks; not production diagnostic PACS replacement.","architectureIntegration":"DICOM→tensor pipeline into Primus 3D ViT + Qwen3.5-4B serving; isolate from clinical write-back.","prerequisites":["GPU memory for 3D tokens","DICOM tooling","License review","IRB dataset"],"constraints":["No regulatory clearance in card","PHI handling complexity"],"security":"Enforce PHI encryption, access logs, and no internet exfiltration from research VPC.","proofOfValue":["Research eval protocol complete","Latency/memory baselines","Radiologist review sample logged"]},"delivery":{"work":"Stand up research environment, train users on non-diagnostic use, run pilot eval.","dependencies":["IRB approval","Imaging IT","Compliance"],"ownership":"Clinical research informatics lead.","skillsAdoption":"Radiologist-in-loop evaluation training.","governanceCheckpoints":["IRB","Security","License acceptance"],"acceptanceCriteria":["Sandbox live on approved data","Non-diagnostic banners present","Eval report filed"],"risks":["Shadow clinical use","PHI breach","Overstated accuracy"]}},"retrievedAt":"2026-09-27T02:58:40Z","enrichedAt":"2026-09-27T03:05:00Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Report-generation tools need accessibility review for clinicians; research pilots should include diverse readers.","procurementImplications":"Procure as research software/weights with explicit non-clinical contract language.","operatingModelImplications":"Keep research model ops separate from clinical production imaging systems.","sourceVerification":{"openedUrl":"https://huggingface.co/nvidia/NV-Reason-CT","referenceExcerpt":"Native 3D vision encoder (Primus 3D ViT, initialized from COLIPRI)","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}