How MyRadAgent compares, on published metrics
The leading AI reporting tools publish their results on their own terms. This page lines those numbers up against MyRadAgent's measured ones, says what each figure actually measures, and links every one to its source. Updated 2026-11-03.
Side by side
| Metric | Rad AI (Omni Impressions / Omni Reporting) | PowerScribe One Smart Impression (Microsoft/Nuance) | MyRadAgent |
|---|---|---|---|
| What it drafts | The impression, from the findings the radiologist has already dictatedVendor product description | The impression, inside the dictation systemMicrosoft support: Smart Impression | The entire report — findings, impression, recommendations, technique, guideline follow-up — from a scratch read, a worksheet or the imagesValidation page |
| Edits after drafting | No edit rate published. 80% fewer words dictated per impressionRad AI case study, Colorado Imaging Associates | 2% of AI impressions needed major changes (impressions only; independent preprint, not peer-reviewed)medRxiv 2024 | 90.3% of full-report drafts signed with no edit; 98.3% of draft text kept (5,269 drafted-then-signed reports; method and exclusions published)Validation page |
| Time saved per report | Impression time 30 s → 21 s (about 9 s per report; one practice's figures)Rad AI case study | "Up to a minute per report" (vendor statement)Healthcare IT News, RSNA | 2.78 minutes per report, estimated — the draft covers every section, not only the impression; basis stated/api/public/stats |
| Draft speed | Not published | Not published | Median 28 seconds for a complete report, measured in productionShift statistics, 2026-11-03 |
| Blinded comparison | Peer-reviewed 2026: radiologists preferred its impressions to those of general-purpose LLMsRad AI announcement, npj Digital Medicine | Not published | Blinded 50-case pilot: the MyRadAgent full-report draft preferred over a general-purpose LLM draft in 44 of 50 (88%). Single reviewer; pre-registered three-reviewer study in progressValidation page |
| Price | Quote only. Third-party estimates: $199–$598 per radiologist per monthusagepricing.com (unconfirmed) | Enterprise quote, plus integrationradiological.ai comparison | $0.35 to $0.25 per report, published, no monthly fee, never more than $399 a month; 14-day free trial, no cardPricing |
| Getting started | Enterprise procurement and PACS/RIS integration | Enterprise deployment inside PowerScribe | A browser, today, with cloud dictation built in. The desktop client writes drafts into PowerScribe on paid plansDesktop client |
| Independent validation | One peer-reviewed paper (2026); earlier conference abstractRad AI announcement | One preprint (2024)medRxiv 2024 | Published method on live data; blinded pilot; pre-registered multi-reviewer study under wayValidation page |
What the numbers mean
- Different job, bigger saving. The leading tools automate the impression after the radiologist has dictated the findings. MyRadAgent automates the report. That is why their own published savings are seconds to a minute per report and MyRadAgent's estimate is 2.78 minutes: the draft covers every section.
- Nine in ten drafts need no edit at all. No impression tool publishes a no-edit rate. The closest published figure, a 2% major-change rate, applies to impressions alone, which are a few lines of a report. MyRadAgent's 90.3% applies to the whole document.
- Blind-tested, same direction. Rad AI's peer-reviewed study and MyRadAgent's pilot both show radiologists preferring the purpose-built draft to a general-purpose LLM's. MyRadAgent's pilot measured that on complete reports, 44 of 50.
- No monthly fee, no procurement. Published pay-per-report pricing against enterprise contracts quoted per radiologist per month, and a trial that starts in a browser instead of an IT project.
What MyRadAgent adds beyond these figures
The table compares what each company has published. These six are not in any published metric, and describe MyRadAgent only.
- Intelligent model routing. MyRadAgent works as a model router built for radiology. Each kind of input (dictated text, technologist worksheets, key images, and CT, MR and ultrasound series) is routed to the AI model selected for it on draft quality, turnaround and cost, as measured in clinical use, rather than relying on a single AI model.
- Drafted from the study, not only the dictation. Key images, technologist worksheets and prior reports are read with your dictated findings, and the measurements are carried through to the report.
- Review Before Signing. Every draft is checked against itself: a finding the impression contradicts, a side that differs between sections, a measurement that changed, or a possible critical result is raised at the top of the report. A clean draft carries no notices.
- The same guideline answer every time. Classification categories and follow-up intervals are written into the impression when the findings call for them, consistently from one report to the next.
- A rapid report in a few seconds, in your wording. Dictate or paste the findings and the complete report returns in a few seconds, in the phrasing and structure of the reports you sign.
- Built for practice in the United States. Drafts go into PowerScribe, Dragon, Fluency or any other reporting system, MIPS documentation is checked on each draft, and business associate agreements are in place.
Read the figures fairly.
- These are different endpoints measured by different parties. Rad AI's and Nuance's figures are their own published results; MyRadAgent's are measured on one radiologist's signed reports with the method published. None of this is a head-to-head trial between the products.
- The MyRadAgent blinded pilot had one reviewer and compared against a general-purpose model, not against Rad AI or PowerScribe. A pre-registered study with three independent reviewers is in progress and will replace the pilot figure.
- Rad AI and PowerScribe integrate deeply with hospital PACS and RIS; MyRadAgent runs alongside them. For a health system that already owns an enterprise dictation platform, that integration is worth real money and this page does not pretend otherwise.
Rad AI, Omni Impressions and Omni Reporting are trademarks of Rad AI, Inc. PowerScribe is a registered trademark of Nuance Communications, Inc., a Microsoft company. MyRadAgent is not affiliated with, sponsored by or endorsed by any of them; the marks identify the products being compared. Competitor figures are reproduced from the linked public sources as of 2026-11-03 and may not reflect current product versions or a specific deployment. Every MyRadAgent output is a draft for review and signature by a licensed radiologist.