AI models can now analyze 3D CT scans and generate written research reports describing anatomical structures and potential findings.
With Nalvera.AI, researchers can use advanced medical imaging models such as Merlin and NV-Reason-CT directly in their browser, without coding, installing specialized software, or owning a local GPU.
In this tutorial, we'll explore the differences between these models and show how to generate a CT research report in four steps.
Merlin vs. NV-Reason-CT: Which Model Should You Choose?
Nalvera.AI provides access to two vision-language models for CT research reporting. Both analyze 3D medical images, but they offer different capabilities.
Merlin: Abdominal CT Analysis
Developed by researchers at Stanford University, Merlin generates organ-by-organ research reports from 3D abdominal CT scans. Beyond reporting, Merlin also supports predicted phenotype findings and exploratory five-year disease-risk estimates.
NV-Reason-CT: Chest and Abdominal CT Reporting
Developed by NVIDIA, NV-Reason-CT supports structured report generation for both chest and abdominal CT scans. It also offers reasoning walkthroughs, which present observations, possible interpretations, and uncertainty in a step-by-step format for researchers to review.
| Feature | Merlin | NV-Reason-CT |
|---|---|---|
| Supported anatomy | Abdomen | Chest and abdomen |
| Structured research reports | Yes | Yes |
| Reasoning walkthroughs | No dedicated task | Yes |
| Additional analysis | Phenotype findings, disease-risk estimates | Region-specific reasoning tasks |
Both models are intended for research rather than clinical diagnosis.
Step 1: Upload Your CT Scan
Start by uploading your CT scan to Nalvera.AI: open Submit Task, choose Captioning as the task type, and add your scan.
Nalvera.AI supports common medical imaging formats, including DICOM and NIfTI, so you can work with existing imaging datasets.
Make sure your scan covers the anatomical region supported by the model you intend to use, and remove patient-identifying information before uploading.
Step 2: Choose Your AI Reporting Model
After uploading your scan, explore the compatible AI models and their available tasks.
For Merlin, select Radiology report (text) to generate an organ-by-organ abdominal CT report, or 5-year disease risk for five-year risk scores for cardiovascular disease, ischemic heart disease, hypertension, diabetes, chronic kidney disease, and osteoporosis.
For NV-Reason-CT, choose from:
- Structured chest report
- Structured abdominal report
- Chest reasoning walkthrough
- Abdominal reasoning walkthrough
For this tutorial, select one of the report-generation tasks.
Step 3: Generate Your Research Report
Review your selected scan, model, and task, along with the displayed processing cost.
Click Submit to queue to start the analysis.
Nalvera.AI handles the processing on managed cloud GPU infrastructure. You can follow the task's progress in the Queue while the model analyzes your scan.
Step 4: Explore and Download Your Results
Once processing is complete, open Results and select your finished task.
The generated report is available in Nalvera.AI's Text Results interface.
Depending on the model and task, the output may contain organ-specific descriptions, structured findings, a numbered impression, or a reasoning walkthrough.
You can review the output directly in Nalvera.AI and download the available text and JSON files for further research.
Important Considerations for Research
AI-generated CT reports can support research into medical image understanding and automated reporting, but their outputs require careful evaluation.
Both models have specific anatomical and technical requirements. For example, Merlin's reporting task focuses on abdominal CT, while NV-Reason-CT processes a selected chest or abdominal region and requires sufficient anatomical coverage for its preprocessing.
AI-generated findings may be incomplete or incorrect. Models can miss abnormalities, describe structures inaccurately, or generate findings not supported by the image. Results should therefore be evaluated against appropriate reference data and reviewed by qualified researchers.
To segment your own scans, see how to run medical imaging AI models online.
Frequently Asked Questions About CT Report Generation
Can AI draft a research report from a 3D CT scan?
Yes, as a research draft. Merlin writes organ-by-organ reports from abdominal CT, and NV-Reason-CT writes structured chest or abdominal reports. Drafts can miss findings or mention findings that are not in the image, so they are not a diagnosis.
Which CT scans can I use?
3D CT, uploaded as a DICOM folder or a .nii.gz file. Merlin reads abdominal CT; NV-Reason-CT reads one 38 cm cube of the chest or abdomen and needs the lungs, or the lung bases, in view. Head CT, MRI and X-rays are not supported.
Can I create reports for many scans at once?
Yes. Add multiple scans in one submission; each scan runs as its own AI task.
Get Started With AI-Powered CT Research Reports
Whether you're exploring organ-level reporting with Merlin or structured CT reports and reasoning with NV-Reason-CT, Nalvera.AI makes it possible to experiment with medical imaging AI without building your own inference pipeline.
Explore the models and choose the one that best fits your research:
Research use only. Nalvera.AI is not a medical device, and AI-generated outputs are not intended for clinical diagnosis or treatment.