Rad AI vs Subtle Medical (2026)
Two radiology products that never touch each other's work. Rad AI writes the report impression from the radiologist's own dictated findings, learning their individual language from prior reports, with oversight built into the workflow itself. Subtle changes the image itself, with deep learning reconstruction letting a department scan faster or at lower dose on existing scanners, backed by eleven clearances and more than twenty five publications. One returns radiologist time, the other returns scanner time, and which matters depends entirely on where your queue actually forms. Note the asymmetry in failure: a degraded report is read and corrected, while a degraded reconstruction produces a cleaner image that lost something nobody will notice.
- It generates the report impression from the radiologist's own dictated findings and learns that radiologist's language from their prior reports.
- The workflow enforces oversight, with the model drafting the impression and the radiologist signing it.
- It publishes specific falsifiable operational figures with named baselines, including follow up completion.
- It changes the image rather than the report, with deep learning reconstruction and denoising letting a department scan faster or at lower dose.
- Eleven clearances across two modalities and more than twenty five peer reviewed publications, with a published standards based integration specification.
- The benefit lands on the scanner rather than the reading list, which is a different constraint and often the binding one.
This comparison is published by AI Health Index, an independent research platform that compares healthcare AI vendors objectively. Rad AI and Subtle Medical are each graded against the same capability taxonomy, from each vendor's own public materials and the regulatory record, under the AI Health Index verification standard. No vendor pays for placement, and no vendor has reviewed this page. How this evidence is graded
Plain facts
| Fact | Rad AI | Subtle Medical |
|---|---|---|
| Primary category | Radiology & Imaging AI | Radiology & Imaging AI |
| Founded | 2018 | Not recorded |
| Headquarters | Berkeley, California | Menlo Park, California |
| Website | radai.com | subtlemedical.com |
Side by Side
Select any grade to read the note behind it.
Product updates
The latest logged changes from each vendor, newest first, each linked to its source.
Subtle Medical
-
Regulatory / FDA
Subtle Medical received FDA clearance for SubtleHD (PET), a next generation AI PET image enhancement product for PET/CT and PET/MR systems.
Source
Each vendor in one paragraph
Written to be quoted whole. Each paragraph opens the AI Health Index profile of the vendor: what the product is, who buys it and what it does.
Generative AI for radiology reporting and follow up. Rad AI Reporting is a cloud native reporting system with real time quality checks before sign off; Rad AI Impressions (Omni Impressions) drafts the report impression from the radiologist's dictated findings, customized to each radiologist's language from the site's de identified historical reports; Omni Unchanged carries stable findings forward from prior reports; Rad AI Continuity tracks follow up recommendations through the EHR and RIS.
Source: AI Health Index, August 2026
Deep learning software that enhances and accelerates image acquisition on existing scanners rather than detecting disease. SubtleHD denoises and sharpens MRI (cleared February 2025, K243250, with a Predetermined Change Control Plan), PET (cleared May 2026, supporting all FDA approved radiotracers) and CT (the company's first CT clearance and its eleventh FDA cleared product), succeeding SubtlePET (2018) and SubtleMR.
Source: AI Health Index, August 2026
Questions buyers ask
How are Rad AI and Subtle Medical priced and sold?
Rad AI sells through agreements with radiology practices and health systems, and inquiries go through a demo request.
With Subtle Medical, no rate card, no pricing basis, no statement of whether the model is per scanner, per site, per study or subscription, no term, no minimum and no volume banding. From the AI Health Index, drawn from each vendor's own published materials and the regulatory record, verified August 4, 2026. No vendor pays for placement.
How do Rad AI and Subtle Medical handle patient data?
In Rad AI's case, the security page describes a de identification pipeline specialized to radiology reports, and the FAQ says the models are trained on de identified historical reports, with no images, and customized for each site from three to five years of its de identified reports.
The SubtleEDGE gateway removes identifiers before images leave the facility, and the cloud environment is encrypted and run as if it held PHI. Under SubtleShare, customers share de-identified images with Subtle after processing, study by study, and the DICOM conformance statement says the program is part of the contract unless the customer opts out. From the AI Health Index, drawn from each vendor's own published materials and the regulatory record, verified August 4, 2026. No vendor pays for placement.
Which settings and specialties do Rad AI and Subtle Medical cover?
With Rad AI, the products cover report generation, impression drafting and follow up of incidental findings, sold to hospital radiology departments, physician owned practices and teleradiology groups; named users include AdventHealth, Advocate Health, Emory, Inland Imaging, LucidHealth and Trinity Health, and four health systems are strategic investors.
In Subtle Medical's case, the products cover MRI of all body parts, PET across tracers, CT, synthetic STIR for the spine and brain alignment, on scanners of any make, in hospitals and imaging centers, including pediatric examples. From the AI Health Index, drawn from each vendor's own published materials and the regulatory record, verified August 4, 2026. No vendor pays for placement.
How much do Rad AI and Subtle Medical leave to the clinician?
The company's FAQ states that a licensed radiologist must always review the AI generated impression, edit it as needed and sign the final report; Rad AI Reporting runs real time quality checks before sign off, and the security page describes human in the loop quality assurance since 2019. The company also states the tool catches clinically significant errors in the radiologist's dictation in about 5 percent of reports.
SubtleHD replaces a noisy fast or low dose acquisition with an enhanced one, and SubtleSYNTH offers the option to skip acquiring a STIR sequence and read a generated one instead; in both cases there may be no unmodified comparator on screen. Denoising can alter or hide small lesions on images that look better, not worse. From the AI Health Index, drawn from each vendor's own published materials and the regulatory record, verified August 4, 2026. No vendor pays for placement.
Related comparisons
Other published head to head assessments involving these vendors or their closest peers.
These change different parts of the imaging pathway and a department may reasonably run both, so this is a sequencing question rather than a choice. Their failure modes are opposite and only one is visible: a reporting assistant that degrades produces text a radiologist reads and corrects, while a reconstruction model that degrades produces a cleaner image that has lost something no reader will notice. Rad AI holds no clearance for any product in its portfolio, and neither vendor publishes subgroup performance or pricing.