Medical scribes in radiology: the real cost and workflow comparison

Some radiology practices still run their reports through a transcription service. A radiologist records findings into a dictation system, a typist converts that audio into text, and the report comes back for review and sign-off. A smaller number of practices go further and employ human medical scribes who type directly during reading sessions.

The appeal of both is real. The radiologist focuses on the images. Someone else handles the words. If the transcriptionist or scribe makes an error, they fix it.

But that division of labour deserves a harder look. Front-end dictation software that processes speech on-device, types at the cursor in real time, and costs a fraction of what a human typist costs per year has changed the calculation. This guide compares the two models honestly, including the cases where a human scribe genuinely wins.

One quick distinction before going further: this post is about human medical scribes and back-end transcription services, not AI ambient scribes. If you are evaluating ambient AI tools versus dictation for radiology reporting, the AI scribe vs dictation comparison covers that territory.

What does a medical scribe do in radiology?

In most clinic settings, a medical scribe sits beside the physician and types during patient encounters. They capture history, physical exam findings, and the assessment and plan as the clinician speaks and works. They function as a real-time typist for the clinical visit.

In diagnostic radiology, the workflow is different. There is usually no patient present during image reading. The radiologist views the study, dictates findings, and produces a written report. A scribe in radiology either sits with the radiologist during a reading session, typing as each report is dictated, or works remotely: the radiologist records audio, and the scribe converts that recording into typed text later.

Both models have real strengths. The radiologist does not need to learn new software or change their dictation habits. Corrections go to the scribe. If the radiologist’s accent or microphone setup makes automated recognition unreliable, a human transcriptionist handles it more consistently. For practices where the back-end transcription workflow is already embedded in department operations, the switching cost to change it is genuine.

Those advantages deserve acknowledgment rather than dismissal. The question is how they compare to what current on-device dictation software can do.

What does a medical scribe cost per year?

The cost of a human scribe adds up quickly and never stops.

A full-time in-person medical scribe in the United States earns approximately $30,000 to $40,000 per year in base salary, based on Bureau of Labor Statistics wage data for medical transcriptionists and closely related roles. Including benefits, training, management overhead, and turnover costs, a full-time scribe realistically costs a practice $45,000 to $55,000 per year in total.

Remote transcription services are cheaper per engagement but still accumulate. Typical pricing runs from seven to 14 cents per transcribed line, or roughly one to two dollars per minute of audio. A radiologist running a full reading list and dictating three to four hours of reports per shift can generate monthly transcription costs in the hundreds to over a thousand dollars. Annualised, that is $5,000 to $15,000 per radiologist per year, growing with reading volume.

None of these costs decrease over time. They continue every year you practice, at whatever rate the service charges at renewal. There is no point at which the transcription bill is paid off.

How does front-end dictation compare on speed and accuracy?

Front-end dictation removes the intermediary entirely. You speak, the text appears in the active report field in real time, and the report is complete before you close the study.

RadMyk processes speech on-device at roughly 220 words per minute, with a radiology-tuned speech model. The published out-of-the-box word accuracy is 96.1%, measured on radiology speech including subspecialty anatomy, laterality, and structured report language. The text appears at the cursor wherever you are working: PACS report fields, RIS text boxes, browser-based reporting tools, Microsoft Word, and Citrix or remote desktop sessions.

The report workflow collapses into a single step. There is no audio to submit, no queue to wait on, and no returned text to review for transcription errors. The radiologist creates and reviews the text in one session.

For comparison: a back-end transcription service introduces a gap between when the radiologist dictates and when the signed report is available. During that window, the report is not signed. For time-sensitive findings or high-volume reading lists, that gap is a real friction. The radiologist also needs to review the returned text for transcription errors, which adds another step to the workflow.

The comparison of medical dictation options for radiologists covers front-end tools in more detail, including where enterprise platforms fit versus owned on-device dictation.

Does patient data stay more private with a scribe or with on-device dictation?

This is a dimension of the comparison that often gets overlooked.

A human scribe or remote transcriptionist hears patient audio that contains protected health information: imaging findings tied to a named patient, clinical indication, relevant history, and the radiologist’s impression. Every person in that documentation chain is a HIPAA obligation to manage. A Business Associate Agreement is required with the transcription service, an audit trail applies to audio sent offsite, and policies need to govern who handles that audio and under what conditions.

On-device dictation does not transmit audio to anyone. The speech model runs on the radiologist’s own machine. There is no transcriptionist, no cloud service, and no vendor receiving patient audio. The HIPAA compliance picture for on-device dictation is straightforward: there is no third party in the dictation path, so there is no BAA needed for the audio processing step.

For radiologists managing data under multiple institutional agreements, including teleradiologists reading for several clients, that simplification matters. Each contracting hospital can be told plainly that patient audio is processed on the radiologist’s machine and never transmitted.

When does a medical scribe still make sense for radiology?

There are real cases where a human scribe is the right answer, and any honest comparison has to name them.

Radiologists with voice limitations, whether from chronic medical conditions, accent-related accuracy issues, or microphone constraints, may find that a skilled human transcriptionist produces fewer errors than automated speech recognition. Guided calibration and a good microphone close much of that gap, but it does not close it entirely for every radiologist in every environment.

Interventional radiology clinics, multidisciplinary meetings, and radiology outpatient consultations can look more like clinic medicine than diagnostic reporting. Where a radiologist is actively talking with a patient, taking history, and documenting a complex encounter, a back-end transcription model or even an AI ambient scribe may fit better than front-end dictation. That is a genuinely different workflow from sitting alone with a PACS worklist.

Practices with deeply embedded transcription workflows also face a real transition cost. If a large team has built quality assurance, coding review, and turnaround time tracking around a back-end transcription model, the first-year overhead of switching is a legitimate objection. The long-run savings argument has to overcome that friction.

When is front-end dictation the better choice?

For diagnostic radiologists doing their own structured reporting, front-end dictation is the right fit in most cases.

The report is produced during the reading session. There is no lag, no queue, and no second person handling patient audio. The radiologist controls what goes into the record and sees it immediately. If there is a recognition error, it is corrected on the spot rather than found later in a returned transcript.

The cost argument compounds over time. A one-time purchase covers years of dictation with no recurring bill. A transcription service or medical scribe costs thousands of dollars per year with no endpoint. For any radiologist who reads a full list across a career, the cumulative difference between the two models runs to tens of thousands of dollars.

Portability is another factor that scribes and transcription services do not solve. A cursor-based dictation tool works in any app the radiologist uses: the PACS at hospital A, the RIS at the outpatient clinic, the browser-based reporting tool at the teleradiology client, and the remote desktop session from home. A transcription workflow is attached to a submission and return process that does not change what reporting interface you are working in.

For radiologists managing the PowerScribe 360 sunset and rethinking their full documentation setup, the PowerScribe alternatives guide covers how the broader workflow changes when the enterprise platform changes.

What does the comparison mean in practice?

The medical scribe model was the most practical answer for radiology documentation for a long time. Sending a recording to a human typist and getting text back worked before software could match that reliability.

Front-end on-device dictation has changed the economics and the architecture. The report is done in real time, audio stays on the machine, and the cost is fixed at one payment rather than compounding every year.

The honest answer is that human scribes and transcription services still win in specific situations: complex outpatient radiology, voice-limitation cases, and departments with established workflows they are not ready to change. For diagnostic radiology reporting by radiologists who own their own workflow, the balance has shifted toward self-dictation.

RadMyk is front-end dictation that runs on-device on macOS Apple Silicon and Windows. It types at the cursor in any app, works offline after setup, and is sold as a one-time payment with no subscription and no renewal.

Radiology trainees use RadMyk free for the full length of their training. Practicing radiologists get a 28-day free trial with no credit card. See trial and pricing details at radmyk.com/pricing.

Voice-to-text is a basic tool of the trade, not a premium you rent forever.