# Radiologist

> Reads images to answer a clinical question — and signs for the answer, including for what nobody thought to ask about.

Page: https://flyvolo.ai/en/careers/radiologist
Impact index: 51/100 · confidence Low confidence · assessed 2026-09-12
This is not a probability of losing a job.

## Scope

Written for diagnostic radiologists reading studies in a hospital or teleradiology setting. Interventional radiology is included as one task but is really a different job with a different exposure. Radiographers and technologists — the people who acquire the images — are a separate occupation with its own page, and their picture is not the same.

## Tasks

### Reading the routine study

- Direction: Automating
- Basis: evidenced (verified records attached)
- Why: This task has the four properties that make anything a strong case for machine work: the input is an image, the output is a label, correctness is checkable against a later outcome, and the training data exists in enormous quantity because every study was already read and reported by someone. It is also the task the 2016 prediction was actually about, and on this task the prediction was directionally right even though its timeline and its conclusion were not.
- What this does NOT establish: Says nothing about how much of a radiologist's day this is, and the share differs enormously between a screening programme and a tertiary hospital. The two records attached here are a count of what regulators have authorised and a count of what training programmes have funded; neither is deployment. A cleared device is not a hospital running it, and a funded residency post is not a study being read — the first is a ceiling on what is permitted, the second a bet on demand years out. What would settle it: a health system's own report of what share of studies are machine-read before a person sees them.
- Records:
  - 2026-09-04 · Capability · FDA's AI-enabled device list reaches 1,614 authorisations, 1,230 of them radiology — https://flyvolo.ai/en/changes/ev-20260904-radiologist-1 (source: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices)
  - 2026-05-26 · Labour impact · US diagnostic radiology training posts rose from 997 to 1,083 across five Matches, filling at 98% or above throughout — https://flyvolo.ai/en/changes/ev-20260526-radiologist-2 (source: https://www.nrmp.org/wp-content/uploads/2026/05/Main_Match_Results_and_Data-2026.pdf)
  - 2016-11-24 · Forecast · Geoffrey Hinton said in 2016 that people should stop training radiologists now, that within five years deep learning would do better, and added that it might be ten — https://flyvolo.ai/en/changes/ev-20161124-radiologist-4 (source: https://www.youtube.com/watch?v=2HMPRXstSvQ)

### Catching what nobody asked about

- Direction: Still human-led
- Basis: evidenced (verified records attached)
- Why: A detection model is trained on the question that was asked. The incidental finding is by definition the answer to a question nobody asked, and it is a large part of what makes reading valuable rather than merely accurate. A known incidental can be made into its own question — at one cancer hospital, AI built for incidental pulmonary embolism cut radiologists' missed rate for it from 44.8% to 2.6% — but each such tool looks for one thing, and US rules require the most common class of radiology AI to leave the full read to the standard of care. The finding nobody has built a model for is still found, or missed, by the person reading the whole scan.
- What this does NOT establish: A judgement about the structure of the work, not a measurement. How often incidental findings change what happens to a patient is not quantified in a way that transfers, and the answer differs by modality and by population. It also does not claim people are good at this — missed incidentals are a known and studied failure.
- Records:
  - 2025-04-01 · Constraint · US rules define radiology AI triage software as software that does not mark the image or remove cases from the reading queue, and runs in parallel with the standard of care, the default for all cases — https://flyvolo.ai/en/changes/ev-20250401-radiologist-7 (source: https://www.govinfo.gov/content/pkg/CFR-2025-title21-vol8/xml/CFR-2025-title21-vol8-sec892-2080.xml)
  - 2020-09-30 · Deployment · At a cancer hospital, AI built to flag incidental pulmonary embolism on routine CT cut radiologists' missed rate for that finding from 44.8% to 2.6%, across 11,736 scans — https://flyvolo.ai/en/changes/ev-20200930-radiologist-8 (source: https://pmc.ncbi.nlm.nih.gov/articles/PMC10141443/)

### Deciding whether to image at all

- Direction: Being augmented
- Basis: evidenced (verified records attached)
- Why: Decision-support tools genuinely help here: appropriateness criteria are rule-shaped and a system can apply them consistently where a busy person will not — though the one national attempt to require them, Medicare's mandate on ordering clinicians, was paused and its rules rescinded in 2023 for operational reasons. What does not transfer is the refusal — declining a study a referring clinician wants is a judgement with a cost that lands on a named person, and the value of the radiologist in that moment is precisely that they can be argued with.
- What this does NOT establish: Nothing here says how often radiologists actually protocol studies rather than rubber-stamping them, and in many settings that step has already been delegated or automated away for reasons unrelated to AI.
- Records:
  - 2023-11-16 · Constraint · Medicare paused and rescinded its rules requiring clinicians to consult decision-support software before ordering advanced imaging, calling the reporting requirement an insurmountable barrier — https://flyvolo.ai/en/changes/ev-20231116-radiologist-6 (source: https://www.govinfo.gov/content/pkg/FR-2023-11-16/html/2023-24184.htm)

### The report somebody acts on

- Direction: Being augmented
- Basis: evidenced (verified records attached)
- Why: Drafting and structuring reports is well-suited to generation and is where most of the visible time saving is claimed. The part that does not move is the calibration: a sentence that hedges too much is useless and one that hedges too little causes an unnecessary operation, and where that line sits depends on this patient, this referrer and this health system.
- What this does NOT establish: The only public figure is the Royal College's census: AI drafting tools are in place in 11% of UK departments, and 24% of those departments' clinical directors say they reduced workload — a judgement, not measured time. Vendor claims about time saved are marketing from a party with a stake in the answer; what would settle it is a health system publishing its own before-and-after turnaround times.
- Records:
  - 2025-09-01 · Deployment · The Royal College of Radiologists' 2025 census, answered by every UK radiology department, found AI tools in 75% of them, and no workload reduction for over three-quarters of clinical directors — https://flyvolo.ai/en/changes/ev-20250901-radiologist-5 (source: https://www.rcr.ac.uk/media/n1fjvrv4/rcr-2025-clinical-radiology-workforce-census-report.pdf)
  - 2016-11-24 · Forecast · Geoffrey Hinton said in 2016 that people should stop training radiologists now, that within five years deep learning would do better, and added that it might be ten — https://flyvolo.ai/en/changes/ev-20161124-radiologist-4 (source: https://www.youtube.com/watch?v=2HMPRXstSvQ)

### Procedures done inside a person

- Direction: Still human-led
- Basis: inferred (VOLO's reasoning, no attached source)
- Why: Physical, irreversible, performed on a person who is present and can be harmed. Robotics in this area assists a hand rather than replacing one, and regulation attaches the act to a licensed individual. This is the half of radiology that the 2016 prediction did not address at all.
- What this does NOT establish: Interventional radiology is really a separate occupation with its own training path, and treating it as one task of this page understates how different it is. It also says nothing about volume, which is where the actual pressure on this work comes from.

### Answering for the machine that read it

- Direction: New task
- Basis: evidenced (verified records attached)
- Why: Work that did not exist before these tools entered the reading room, and it lands here because the signature does. A model validated elsewhere is not validated here — performance moves with scanner, protocol and population — so somebody local has to check, and checking is a radiologist's skill applied to a machine instead of to a patient.
- What this does NOT establish: New work appearing is not new headcount, and in most departments this is absorbed by whoever is already there. What would settle it is a hospital publishing a post created for this and the hours attached to it.
- Records:
  - 2026-09-04 · Capability · FDA's AI-enabled device list reaches 1,614 authorisations, 1,230 of them radiology — https://flyvolo.ai/en/changes/ev-20260904-radiologist-1 (source: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices)
  - 2026-07-21 · Constraint · Korea's AI Framework Act places the use of AI medical devices inside high-impact AI, obliging the user to keep it under human management and supervision and to be able to explain its output — https://flyvolo.ai/en/changes/ev-20260721-radiologist-3 (source: https://www.law.go.kr/%EB%B2%95%EB%A0%B9/%EC%9D%B8%EA%B3%B5%EC%A7%80%EB%8A%A5%EB%B0%9C%EC%A0%84%EA%B3%BC%EC%8B%A0%EB%A2%B0%EA%B8%B0%EB%B0%98%EC%A1%B0%EC%84%B1%EB%93%B1%EC%97%90%EA%B4%80%ED%95%9C%EA%B8%B0%EB%B3%B8%EB%B2%95)
  - 2025-09-01 · Deployment · The Royal College of Radiologists' 2025 census, answered by every UK radiology department, found AI tools in 75% of them, and no workload reduction for over three-quarters of clinical directors — https://flyvolo.ai/en/changes/ev-20250901-radiologist-5 (source: https://www.rcr.ac.uk/media/n1fjvrv4/rcr-2025-clinical-radiology-workforce-census-report.pdf)

## How to cite this

- The unit is the task, not the job title: say which tasks are changing and which are not.
- Carry each task's basis and its limits with it; an inferred judgement is not a finding.
- This page gives no number of years. Do not derive one from it.
- More: https://flyvolo.ai/llms.txt