For twenty years, “radiology voice recognition software” meant one thing: a speech engine that turned a radiologist's words into text faster than typing. That category still exists and still matters — but since 2023 it has been joined by two genuinely new architectures. Generative AI now drafts impressions and whole report sections from terse dictation. And voice-to-report pipelines go further still: the dictation comes back as a structured, template-formatted draft that has already been checked against what was actually said.
This guide maps all three architectures and the notable products in each, worldwide. Full disclosure up front: we build one of them — Bionic Flow, 5C Network's voice-to-report product, currently in beta. Every other vendor here is described from its own published materials, accuracy numbers are labelled self-reported for all vendors including us, and where a competitor is the better fit we say so. If you are specifically comparing within India, the companion guide to radiology reporting software in India covers RIS-PACS and reporting-as-a-service as well.
These are three different machines wearing one name
Before comparing vendors, decide which question you are actually answering:
- Speech recognition (transcription layer). Converts your words to text in real time, inside a reporting screen you already own. You structure, check and finalise the report yourself. The right buy when typing speed is the only bottleneck and your workflow is otherwise dialled in. Products: Dragon Medical One, Augnito, and the front-end engine inside Fluency for Imaging.
- Reporting platforms with speech inside. The reporting workflow itself — worklist, templates, structured reporting, PACS/RIS/EHR integration — with speech recognition embedded. An enterprise purchase and an enterprise integration project. Products: PowerScribe One, Solventum Fluency for Imaging.
- Generative-AI drafting and voice-to-report pipelines. The newest layer: AI that writes parts of the report (impressions, carried-forward findings, expanded prose) or structures the entire draft from dictation. The productivity ceiling is highest here — and so is the question every buyer must ask: what stops the model from writing something the radiologist never said? Products: Rad AI Omni Reporting, New Lantern's Curie, Bionic Flow.
Two cross-cutting variables apply to all three: deployment (nearly everything here is cloud-first; Fluency for Imaging is notable for offering on-premises as well as cloud, per Solventum) and integration reality — whether the tool types into any screen, requires named PACS/RIS integrations, or deliberately avoids integration altogether.
The tools, machine by machine
Descriptions below are compiled from each vendor's published materials and public reporting as of August 2026. Verify current capabilities, availability in your country, and pricing directly with the vendor — this market is moving fast.
Speech recognition (transcription layers)
Dragon Medical One (Microsoft / Nuance)
The global default for medical dictation. Dragon Medical One is Nuance's cloud-based speech-recognition product, hosted on Microsoft Azure since Microsoft's 2022 acquisition of Nuance, and offered with radiology-specific terminology, vocabulary and formatting. Microsoft's materials claim up to 99% accuracy with no voice-profile training, automatic accent detection, and documentation up to three times faster than typing (all self-reported). It types into whatever application has focus, which makes it the pragmatic choice for radiologists who want best-in-class recognition inside an existing non-Nuance workflow. Sold per-user by subscription, including through the Microsoft marketplace.
Augnito (Scribetech)
Indian-built, cloud-based medical speech recognition from Scribetech (Mumbai, with UK roots dating to 2001 serving NHS transcription), with a dedicated radiology specialty. Augnito's materials claim 99% out-of-box accuracy with no voice enrollment, and describe embedded integrations with PACS/RIS vendors including Intelerad, Fujifilm Healthcare UK, Magentus and OpenRad, availability in roughly half of NHS radiology departments, and general availability in India and the Gulf since 2020. A strong choice where accent diversity matters and where your PACS vendor already embeds it. We compare it with our own product directly in Bionic Flow vs Augnito.
Reporting platforms with speech inside
PowerScribe One (Microsoft / Nuance)
The incumbent giant of radiology reporting. Nuance's materials describe the PowerScribe platform as used by more than 80% of US radiologists, and PowerScribe One — introduced in 2018 — converts free-form dictation into structured data, integrates AI algorithms into the reading workflow, and added a generative-AI auto-impression capability in 2023 (per Nuance's announcements). It is the enterprise standard for deep PACS and EHR integration, with the enterprise implementation timeline to match. One current buyer consideration, per industry reporting in 2026: Microsoft is retiring the older PowerScribe 360, so sites on the legacy product face a migration decision — which is exactly the moment to re-evaluate the whole category rather than default-upgrade.
Solventum Fluency for Imaging (formerly 3M M*Modal)
The perennial challenger to PowerScribe, now under Solventum after 3M's healthcare spin-off. Fluency for Imaging combines front-end speech recognition with natural-language understanding that contextually interprets the dictation, plus computer-assisted documentation prompts inside the reporting workflow. Per Solventum, it has ranked #1 in Best in KLAS for front-end imaging speech recognition in 2025 and multiple prior years, is interoperable with leading PACS, RIS and EHR systems, and — unusually in this cloud-first field — is offered both on-premises and in the cloud. The natural shortlist entry for enterprise buyers who want a credible alternative quote against Nuance.
Generative-AI drafting and voice-to-report pipelines
Rad AI Omni Reporting
The most visible of the generative-AI-native reporting companies. Rad AI's Omni Reporting drafts the impression automatically, carries unchanged prior findings forward so the radiologist dictates only what is new, and expands terse dictation into complete report language — integrating with existing reporting systems rather than replacing them. Rad AI's materials claim up to 90% fewer dictated words, efficiency gains up to 50%, and clinically significant error catches in about 5% of reports (all self-reported), with the radiologist reviewing and confirming every AI-assisted report before signing. Primarily a US-market product; strongest where a group already runs PowerScribe or similar and wants an AI layer on top.
New Lantern (Curie)
A San Francisco startup (founded 2021, Series A led by Benchmark) rebuilding the whole stack around AI: Curie combines a smart worklist, cloud PACS viewer and an LLM-based AI reporter in one platform. Per New Lantern, the system reads technologist worksheets via OCR, lets radiologists free-dictate positive findings without commands or punctuation, weaves them into complete sentences in the radiologist's own template, and generates guideline-based impressions — with a claimed automation of over 75% of non-diagnostic work (self-reported). The bet is platform replacement rather than augmentation, which makes it most relevant to groups willing to change PACS and reporting together.
Bionic Flow (5C Network) — our product
We build this one, so read accordingly. Bionic Flow is voice-to-structured-report: dictation is not transcribed into a box but mapped into the radiologist's own templates and macros during dictation, returning a structured, sign-ready draft. The architectural distinction is the validation layer — clinical AI checks audit the draft against what was actually said, flagging flipped laterality, missed measurements, negation errors and internal inconsistencies before sign-off, and content that cannot be traced to the radiologist's voice, macros or templates is blocked outright. It is deliberately RIS/PACS-agnostic (the validated draft is copied and pasted into any system, no integration project), only dictation audio leaves the institution (discarded by default), no voice enrollment is needed, and it covers 23 radiology subspecialties. Honest status: Bionic Flow is in beta — it is how radiologists on 5C Network's own reporting platform dictate today, and external access is through a public waitlist. Pricing is per radiologist per month, volume pricing above 10 seats, 30-day opt-out.
This is not an exhaustive list — the field also includes regional speech vendors, EHR-bundled dictation, and a fast-growing crop of LLM reporting startups. We have focused on the names most often shortlisted globally in 2026.
Buy for your bottleneck
- Typing speed is the bottleneck; the workflow is fine. Buy a transcription layer — Dragon Medical One globally, Augnito where its PACS integrations or accent handling fit. Cheapest intervention, fastest payback.
- You are buying or replacing the reporting workflow itself. Shortlist PowerScribe One and Fluency for Imaging, budget for an enterprise integration project, and get migration terms in writing — especially if you are being moved off a legacy product.
- Your radiologists spend their time structuring and double-checking, not typing. Evaluate the generative layer: Rad AI Omni if you want AI on top of an existing reporting system, New Lantern if you are willing to replace the stack, Bionic Flow if you want the structuring and validation done during dictation with no integration project — and can work with a beta via waitlist.
- Whichever category: trial with your own dictations and accents, ask for the anti-hallucination mechanism in writing, confirm what patient data leaves your network, and compare cost per radiologist per year against minutes genuinely saved per report.
Frequently asked questions
Is AI dictation accurate enough for radiology?
For transcription, yes — with the caveat that every accuracy figure you will read is vendor self-reported, including ours. Microsoft's materials claim up to 99% accuracy for Dragon Medical One with no voice-profile training; Augnito's materials claim 99% out of the box; Solventum cites top Best in KLAS rankings for its front-end imaging speech recognition. Real-world accuracy depends on your accent, your microphone, your reading-room noise and your subspecialty vocabulary, which is why no published number substitutes for a trial. The practical test: dictate twenty of your own reports, in your own voice, with your own templates, and count the corrections. Accuracy of transcription is also not the whole question — a perfectly transcribed report can still contain a dictated error, which is why the newer generation of tools adds validation on top of recognition.
Does Dragon work for radiology dictation?
Yes. Dragon Medical One, Nuance's cloud speech-recognition product now owned by Microsoft, is offered with radiology-specific terminology, vocabulary and formatting, and is widely used by radiologists dictating into third-party reporting screens. The distinction to understand is that Dragon Medical One is the speech engine, while PowerScribe One is Nuance's dedicated radiology reporting platform built around that engine — with structured reporting, worklist integration and PACS/EHR connectivity. If you want dictation inside a reporting workflow you already own, Dragon Medical One is the relevant product; if you are buying the reporting workflow itself, PowerScribe One is the incumbent — Nuance's materials describe it as used by more than 80% of US radiologists.
What is the hallucination risk with AI-generated radiology reports?
Real, measurable, and the single most important question to ask any generative-AI reporting vendor. Peer-reviewed work in 2024–2026 has documented hallucinations — plausible but incorrect content such as phantom findings, wrong laterality or fabricated measurements — in LLM outputs for medical imaging, with published analyses reporting rates in the high single digits to mid-teens for unconstrained systems; research frameworks like ReXTrust exist precisely because the problem is common enough to need automated detection. The honest mitigation is architectural: constrain what the model may write and validate the draft against a ground truth. Bionic Flow's approach — and we build it, so weigh that — is to block any content that cannot be traced to the radiologist's own voice, macros or templates, and to run clinical AI checks comparing the draft against the dictation for laterality, measurements, negation and consistency before signing. Whatever tool you evaluate, ask the vendor to show you, in writing, what prevents the draft from containing something the radiologist never said.
How are radiology dictation and AI reporting tools priced?
Almost universally by per-user subscription, and almost universally by quotation rather than a public price list. Dragon Medical One is sold as a per-user cloud subscription (including via the Microsoft marketplace); PowerScribe One and Solventum Fluency for Imaging are enterprise deployments priced by quotation, typically with implementation and integration costs on top; Augnito is a per-user subscription with rates by quotation; Rad AI and New Lantern price by quotation as well. Bionic Flow is priced per radiologist per month, with volume pricing above 10 seats and a 30-day opt-out. The comparison that matters is cost per radiologist per year against minutes actually saved per report — a cheaper tool that saves two minutes per case loses to a costlier one that saves eight — plus the integration project cost, which for enterprise platforms can exceed the licence.
Will these tools integrate with my existing PACS and RIS?
It depends on the category. Speech-recognition layers like Dragon Medical One and Augnito type into whatever reporting screen has focus, and both also offer embedded integrations — Augnito's materials list Intelerad, Fujifilm (UK), Magentus and OpenRad among its PACS/RIS partners. Reporting platforms like PowerScribe One and Fluency for Imaging are the integration: they connect to your PACS, RIS and EHR as an enterprise project measured in months. Generative assistants like Rad AI Omni integrate with existing reporting systems, including PowerScribe. Bionic Flow takes the opposite bet: it is deliberately RIS/PACS-agnostic — the validated draft is copied to your clipboard and pasted into whatever system you already run, with no integration project, and deeper hooks available on 5C's own RIS. Ask every vendor the same two questions: which named systems have live embedded integrations, and what happens on day one if mine is not on the list.
Bionic Flow turns dictation into a structured, validated draft — checked for laterality, measurements and negation against what you actually said — with no RIS/PACS integration project. It is in beta; leave your details and we'll set up a walkthrough on your own studies, right from this page.
See Bionic Flow on your own studies
Tell us where to reach you — a 5C radiology consultant responds within one business day.
Prefer the product page? Join the Bionic Flow waitlist or compare Bionic Flow vs Augnito.
