Radiology dictation software: how to choose in 2026

Radiology dictation software is sold under one name but covers three different products: front-end dictation that types what you say, enterprise reporting platforms that happen to include a speech engine, and ambient AI scribes that listen to a conversation and draft a note. A radiologist reading a CT at 9 pm needs the first one. Most search results push the second or third.

This guide is for the radiologist choosing for themselves, or the group lead choosing for a reading room. It covers what the software has to do, how accuracy numbers are produced, why cloud versus on-device matters more in radiology than in a clinic, whether it works with your PACS, what it costs over three years, and how to test a tool in a week. RadMyk is one of the options discussed, and the sections say where it does not fit as plainly as where it does.

Key takeaways

• Radiology dictation is front-end dictation: speak, and text lands in the report field. Enterprise reporting platforms and ambient scribes solve different problems.
• Accuracy figures are only comparable when the test conditions are published. The only open radiology benchmark we know of measures 96.1% word accuracy out of the box.
• On-device processing means audio never leaves the reading room and dictation keeps working when the network does not.
• Most options are per-user subscriptions. The one-time alternative costs less than a year of the cheapest published subscription.
• Test with your own voice, your own microphone, and your own PACS before you decide. A week is enough.

What does radiology dictation software do?

Radiology dictation software turns a radiologist’s speech into report text while the images are open. The output goes into a Picture Archiving and Communication System (PACS) or Radiology Information System (RIS) report field, a browser-based reporting tool, an EHR text box, or a Word document. The job is narrow: hear “no focal consolidation, effusion, or pneumothorax” and type exactly that, at the cursor, at reading speed.

Three product categories claim the name.

Front-end dictation is the direct replacement for typing. You speak, the words appear, you correct and sign. Nuance Dragon Medical One, Dolbey Fusion Narrate, Augnito Spectra, and RadMyk are in this category.

Enterprise reporting platforms include speech recognition as one part of a larger system. Nuance PowerScribe and M*Modal Fluency for Imaging add structured templates, peer review, critical findings management, quality checks, and PACS/RIS integration. The speech engine is not the product. The platform is.

Ambient AI scribes listen to a clinician-patient conversation and draft a note afterwards. Nuance DAX, Suki, and Heidi do this. Radiology rarely has a conversation to summarize, which is why an AI scribe and radiology dictation solve different problems.

Most individual radiologists, teleradiologists, locums, and small groups need front-end dictation. A hospital department that needs peer review workflow and structured encoding needs a platform, and should buy one. The mistake is buying a platform when the requirement was accurate text in a field, or buying a general clinical dictation tool when the requirement was radiology vocabulary.

How accurate is radiology dictation software?

Accurate enough to use for every report, if the tool has radiology vocabulary and you can see how the number was measured. Vendors quote figures between 95% and 99% word accuracy. Those numbers are not comparable, because the test conditions almost never appear next to them.

Word accuracy is 100% minus word error rate. A 96% figure means four words in a hundred need correction. In a 180-word chest CT report, that is about seven corrections. Whether that is acceptable depends on which words are wrong. A missed “no” in front of “pneumothorax” is a different error from a dropped “the”.

Three things move the number more than the engine does. The microphone and room: a wired headset in a quiet reading room and a laptop microphone in an open-plan office produce different error rates from the same software. The speaker: accents, speaking pace, and dictation habits all matter, which is why tools with a short calibration step gain a point or two after setup. The test material: a vendor measuring on its own curated audio will beat the same vendor measured on your Monday morning worklist.

The useful question to ask a vendor is not “what is your accuracy” but “measured on what, by whom, and can I hear the clips”. RadMyk publishes an open radiology benchmark with every test clip and every error playable on the page. The result is 96.1% word accuracy out of the box, before calibration, at roughly 220 words per minute. You can disagree with the methodology, but you can see it. Our guide to radiology dictation accuracy covers how to read vendor claims in more detail.

Does it need radiology vocabulary, or will general medical do?

It needs radiology vocabulary. General medical speech recognition handles “hypertension” and “metformin” well and stumbles on “subpleural”, “spiculated”, “hypoattenuating”, and “BI-RADS 4A”. The error rate on specialist terms is where general models fall over, and radiology reports are dense with them.

Vocabulary is also about cadence. Radiology reports have a predictable structure: technique, comparison, findings by organ system, impression. A model trained on that structure predicts “No evidence of” after “Lungs:” and “unremarkable” after “Adrenal glands are”. A general model does not, and the difference shows up as corrections.

This is why the Nuance family splits into two products. Dragon Medical One is general clinical dictation. PowerScribe is where Nuance puts its radiology depth. A radiologist buying Dragon Medical One alone gets a strong engine with a vocabulary gap, which the RadMyk vs Dragon Medical One comparison goes through line by line. Augnito and Dolbey both offer radiology models, Augnito as one of 55+ specialty models and Dolbey through its separate Fusion Narrate Dx product. RadMyk’s single model is radiology-tuned from the start.

Measurement conventions matter too. “1.2 by 0.8 centimeters” should become “1.2 x 0.8 cm” without a macro. Laterality, series and image numbers, and Hounsfield units are all places where a general engine needs correcting and a radiology engine does not.

Cloud or on-device: what happens when the network drops?

Cloud dictation stops. On-device dictation keeps going. That one sentence is most of the argument, but the privacy side matters as much.

Cloud dictation sends your audio to a vendor server, which returns text. Dragon Medical One and PowerScribe One stream to Microsoft Azure. Dolbey Fusion Narrate streams to nVoq, a third-party engine. Augnito’s standard product streams to Augnito’s cloud, with an on-prem option for stricter environments. The vendor needs roughly 4 Mbps down and 2 Mbps up for Dragon Medical One, and the connection has to hold for the whole session.

Reading rooms are not clinics. They sit behind hospital firewalls, on VPNs, in basements, and in teleradiology home offices on consumer broadband. A dropped VPN during a night shift means a stopped report. Our guide to offline medical dictation software covers the failure modes in detail.

On-device dictation runs the speech model on the computer in front of you. After a one-time engine download, no audio leaves the machine and no network is needed. Privacy is a property of the architecture rather than a clause in a business associate agreement. There is no vendor server holding patient audio, because there is no vendor server.

The honest trade-off: cloud tools can update their models continuously and offer portable profiles that follow a clinician between hospitals. Dragon Medical One is good at this. For a radiologist who reads in one or two places and cares about audio staying in the room, on-device wins. For a teleradiology group, offline operation is the difference between finishing the worklist and emailing an apology, which is why teleradiology dictation software has its own set of requirements.

Does it work with my PACS, RIS, and EHR?

The answer depends on whether the tool integrates or types. Integration means a vendor connector into a specific system. Typing means the software inserts text wherever the cursor is, the way a keyboard does.

PowerScribe and Fluency for Imaging integrate deeply with PACS and RIS, and that integration is a large part of what a department pays for. It brings worklist context, structured fields, and report routing. It also means an implementation project, a vendor contract, and dictation that works inside the integrated system and nowhere else.

Dragon Medical One integrates best with the EHRs it has connectors for, Epic, Oracle Cerner, and MEDITECH among them. Dolbey has tested against roughly 100 EHRs and is strong on shortcut automation inside them.

Cursor-typing tools work in any application that accepts typed text. That includes every PACS and RIS report field, browser-based reporting tools, Word, Notes, and Citrix or remote desktop sessions, because the text is typed into the session the same way a keyboard would type it. RadMyk works this way. There is no integration project because there is nothing to integrate. There is also no worklist context and no structured encoding, so if your department needs those, a typing tool is the wrong category.

For a radiologist who reads across a hospital PACS, a satellite site’s different PACS, a browser tool when covering remotely, and Word for teaching files, cursor typing is the only model that follows you to all four.

Does radiology dictation software run on a Mac?

Mostly no, which is a long-standing gap. PowerScribe has no Mac client. Dragon Medical One has no native Mac application; running it on a Mac means Parallels or Boot Camp and a maintained Windows environment alongside macOS. Fluency for Imaging is Windows and web, IT-managed. Dolbey is Windows-first. Augnito has web and mobile access but no native desktop Mac app.

RadMyk runs natively on Apple Silicon (M1 and later) and on Windows. For radiologists reading from home on a MacBook Pro, or in a private practice that standardized on Macs, that is the whole decision. The medical dictation software for Mac guide goes through the virtualization workarounds and why they are worse than they sound. Windows users have the opposite problem, too many options and most of them subscriptions, covered in the Windows medical dictation guide.

What does radiology dictation software cost?

Most of it is a per-user subscription, and most of the prices are not published. Here is what is.

Dolbey Fusion Narrate publishes $850 per user per year, with a $500 one-time onboarding fee and a $350 per year AI Assist add-on. Dragon Medical One is not published officially; reseller-cited figures run about $79 to $99 per provider per month on multi-year terms, plus an implementation fee reseller-cited at around $525. PowerScribe, Fluency for Imaging, and Augnito are quote-based and negotiated per organization. Augnito third-party estimates range widely and should be treated as directional only.

RadMyk is a one-time purchase. The first 100 users pay $199 (₹14,999 in India), after which the price reverts to $299. One license installs on two devices with one active at a time. Future updates are included. There is no renewal, no per-report meter, and no seat fee.

Three-year math, using only the published and reseller-cited figures above:

Tool	Year 1	3 years
Dolbey Fusion Narrate (published)	$1,350	$3,050
Dragon Medical One (reseller-cited, $79 to $99/month plus setup)	$1,473 to $1,713	$3,369 to $4,089
RadMyk (first-100 price)	$199	$199

The subscription model exists for a reason. Cloud processing costs the vendor money every month, so they bill every month. On-device processing moves that cost onto hardware you already own, which is what makes a one-time price possible. The pricing page has the full feature list and the lifetime-cost table. For departments, the one-time price books as a capital purchase rather than recurring operating expense, and licenses can be granted and revoked centrally. The departments page covers rollout and procurement.

Which radiology dictation software should you consider?

Six products cover the real market. Each one is the right answer for someone.

Nuance PowerScribe is the incumbent enterprise radiology reporting platform in the United States. It wins on structured reporting, RSNA and ACR Common Data Elements, Smart Impression drafting, in-workflow quality checks, peer review, and PACS/RIS integration. It is cloud-based in its current PowerScribe One form, has no Mac client, and is sold on enterprise contracts. PowerScribe 360 support ended August 31, 2026, with full end-of-life on August 31, 2027, which is forcing a lot of departments to decide now; the PowerScribe 360 end-of-life guide lays out the options. The RadMyk vs PowerScribe comparison is the long version.

M*Modal Fluency for Imaging is the other enterprise platform, with CAPD nudges, actionable findings management, structured encoding, and peer review. It has been Best in KLAS for front-end imaging speech several times. It is mid-divestiture from Solventum to Jacobian, which is a buyer risk worth pricing in. See RadMyk vs M*Modal Fluency for Imaging.

Nuance Dragon Medical One is the best general clinical cloud dictation product, Best in KLAS six years running. Portable profiles, no voice training, automatic accent detection, and deep EHR integration are genuine strengths. It is general medical rather than radiology-specific, cloud-only, subscription-only, and has no native Mac client.

Augnito is one of the stronger modern cloud voice-AI suites, with 55+ specialty models including radiology, an ambient scribe product, mobile apps, browser extensions, and a strong presence in the UK, India, and the Middle East. It is a per-user subscription, cloud by default, with no native Mac desktop app. See RadMyk vs Augnito.

Dolbey Fusion Narrate has the most transparent pricing in the category and the best command automation, with image-based “Vision Click” commands, multi-step shortcuts, and scripting for power users. It wraps the third-party nVoq cloud engine, is Windows-first, and splits radiology into a separate on-prem Fusion Narrate Dx product. See RadMyk vs Dolbey Fusion Narrate.

RadMyk is front-end radiology dictation that runs on-device, works offline after setup, types at the cursor in any application, runs natively on Apple Silicon and Windows, and is sold once. It has radiology-tuned vocabulary, 96.1% measured word accuracy, a guided calibration step, voice macros, and an optional webhook for teams. It has no peer review, no structured encoding, no worklist, no EHR connectors, and no ambient scribing. It is the right choice for the radiologist who wants to own the dictation layer and the wrong choice for the department that is buying a reporting platform.

The column where RadMyk is alone: one-time payment, on-device, native Mac, and any-app typing in one product. Each competitor has at most two of the four.

Is radiology dictation software free for residents?

RadMyk is. Residents and fellows use it free for the whole of their training, with proof of residency, and get a 30-day bridge after qualifying. There is no countdown and no credit card. The free for trainees post explains the reasoning, which is partly generous and partly self-interested: trainees dictate a lot and tell us what breaks.

None of the other five products listed above has a free tier for trainees. Residents in most departments use whatever the department licensed, which is usually PowerScribe, and have no dictation at home.

For everyone else, RadMyk has a flat 28-day free trial with no credit card and no email required. Download it, calibrate, and dictate into your real worklist for four weeks.

How do you test radiology dictation software in a week?

Test with your own voice, your own microphone, your own reports, and your own PACS. A vendor demo tells you how the vendor dictates.

Day one, set up and calibrate. Use the microphone you will use in practice, not a better one. If you are choosing a microphone at the same time, the microphone guide for medical dictation covers headsets, desktop mics, and handhelds for reading rooms.

Days two to four, dictate real reports in your real report field. Count corrections per report across at least 30 reports and note which words fail. Specialist terms, measurements, laterality, and negations are the ones that matter. A tool that gets “pneumothorax” wrong once in 30 reports is in a different class from one that gets “no” wrong once in 30 reports.

Day five, break the network. Unplug the cable or turn off wifi mid-report and see what happens. If the tool stops, that is what your night shift looks like when the VPN drops.

Day six, try the places you read that are not your main workstation: the satellite PACS, the browser tool, the Citrix session, the Mac at home. A tool that only works in one of them is a tool for one of them.

Day seven, do the three-year math with the figures the vendor will put in writing, including implementation and add-on fees. Then decide.

For a wider framing of the whole category, including transcription services and ambient scribes, the medical dictation software buyer’s guide covers the non-radiology options, and our guides to radiology voice recognition software and medical speech recognition software for radiology go deeper on vocabulary and engine evaluation.

Frequently asked questions

What is the best radiology dictation software?

There is no single best. PowerScribe is the best enterprise radiology reporting platform. Dragon Medical One is the best general clinical cloud dictation product. RadMyk is the best fit for a radiologist who wants on-device, offline, Mac and Windows dictation bought once rather than rented. Pick the category first, then the product.

Does radiology dictation software work offline?

Cloud products such as Dragon Medical One, PowerScribe One, Augnito cloud, and Dolbey Fusion Narrate do not; they need a live connection to the vendor’s servers. On-device products such as RadMyk work offline after a one-time engine download, because the speech model runs on your own computer.

Does radiology dictation software work on a Mac?

PowerScribe, Dragon Medical One, Fluency for Imaging, and Dolbey have no native Mac client. RadMyk runs natively on Apple Silicon Macs and on Windows.

How accurate is radiology dictation software?

Vendor figures run from 95% to 99% word accuracy, but the measurement conditions are rarely published. RadMyk measures 96.1% word accuracy out of the box on an open benchmark where every clip and error is playable. Your result depends on your microphone, your room, and your speaking pace, which is why you should test with your own setup.

How much does radiology dictation software cost?

Most products are per-user subscriptions. Dolbey publishes $850 per user per year plus $500 onboarding. Dragon Medical One is reseller-cited at $79 to $99 per provider per month. PowerScribe, Fluency, and Augnito are quote-based. RadMyk is $199 one-time for the first 100 users, reverting to $299, with no renewal.

Is there radiology dictation software that is free for residents?

RadMyk is free for radiology residents and fellows for the length of their training, with a 30-day bridge after qualifying. Everyone else gets a 28-day free trial with no credit card.

The bottom line

Radiology dictation software should put accurate text at your cursor, in whatever system you are reading in, without sending your voice anywhere and without stopping when the network does. If your department is buying a reporting platform with peer review and structured encoding, buy PowerScribe or Fluency and judge them as platforms. If you are a radiologist, a teleradiologist, or a group that wants to own the dictation layer, the requirements are radiology vocabulary, on-device processing, Mac and Windows support, and a price you pay once.

RadMyk is built to that list. The trial is 28 days with no card, trainees use it free, and the benchmark is open. Dictate your own worklist for a week and count the corrections. That number is the only one that matters.