EXPOSED
Image interpretation is the single most attacked task in medicine by AI, and detection models already match or beat human sensitivity on narrow tasks like lung nodules, fractures, and mammographic screening — but the modal radiologist also compares priors across years, resolves ambiguous findings against clinical history, dictates actionable reports, fields curbside consults from surgeons and oncologists, and performs image-guided procedures. What holds is the signature: a licensed, board-certified human owns the report and the malpractice exposure, and no vendor currently accepts autonomous diagnostic liability. Expect volume per radiologist to rise sharply while the interpretive commodity tier compresses.
Most of this decline happened after 2021 — it is not the pandemic dip.
This line is counted by the Bureau of Labor Statistics — the one figure on this page that isn't a judgement of ours. Headcount moves on demand, offshoring, demographics and the business cycle, and automation is one term among several, often not the loudest.
So a falling line is not evidence that AI did it, and a rising one is not evidence that it won't. Both happen in this register: some occupations resist automation and shrink anyway, others are highly automatable and keep growing.
BLS projection, 2024–2034
+2.7% 28,200 → 29,000 on the projections basis
Growing, and only partly exposed
The BLS expects +2.7% more of these jobs by 2034, and at 56/100 the work is only partly exposed — some tasks are automatable, the core of the job is not. Nothing here is in tension.
Different clocks. The score is what current AI could do to this work today. The projection is how many of these jobs will exist in 2034. Everything between the two — how fast employers actually adopt, whether demand grows in the meantime — is why they can point opposite ways without either being wrong.
~800 openings a year on average, including replacing people who leave.
PhysicianRadiologistMammographerTeleradiologistNeuroradiologistNuclear PhysicianAttending PhysicianGeneral RadiologistNuclear RadiologistRadiology PhysicianVascular RadiologistAttending RadiologistPediatric RadiologistRadiologist PhysicianDiagnostic RadiologistVeterinary RadiologistTherapeutic RadiologistNuclear Medicine OfficerBreast Imaging RadiologistInterventional RadiologistNuclear Medicine PhysicianNuclear Medicine SpecialistInterventional NeuroradiologistInterventional Radiology Physician
Holding it up: liability shield . Weakest point: trust premium .
Mixed — a routine tier and a judgment tier An 8 reflects that the highest-volume portion of the workday — screening mammography reads, chest CT nodule detection, plain-film fracture triage, bone-age and cardiothoracic ratio measurement — is exactly where FDA-cleared CAD and triage algorithms already operate at scale, but it stays above the 6 line because worklists still contain multi-phase abdominal CTs where the finding only makes sense against three years of priors and a surgical note, plus fluoroscopy, biopsies, drain placements and tumour boards that no model touches.
Some physical or field component Most of the day is a dark reading room and a PACS workstation, which anchors this low, but the 8 accounts for the interventional and procedural share of a general radiologist's practice — CT- and ultrasound-guided core biopsies, thoracentesis, paracentesis, lumbar punctures, fluoroscopic barium studies, arthrograms — performed on a live patient in a controlled suite rather than the field.
Licensed human required and personally liable An 18 is earned by state medical licensure plus ABR board certification, hospital credentialing and privileging per procedure, and the fact that the dictated report carries the radiologist's electronic signature and is the discoverable document in missed-cancer litigation — the discount from 20 is that most reads are one node in a care chain where the ordering clinician and referring physician also carry duty, so radiology malpractice exposure is real but rarely sole.
Exists to be accountable for ambiguous calls A 15 fits calls where the imaging is genuinely equivocal and the recommendation has consequences: BI-RADS 3 versus 4 on an indeterminate mass, whether a subtle hyperdensity on a stroke code is haemorrhage or artefact within the thrombolysis window, calling an unexpected pulmonary embolism at 3am and escalating it, or recommending against biopsy on a lesion the surgeon wants sampled — ACR appropriateness criteria and structured reporting templates constrain the wording but not the decision.
The verdict above describes this occupation as a whole. Almost nobody does the typical version of a job — tick what's actually in your week and see how your own mix sits.
Your task mix speaks to task resistance (8/20 here) — how much of the day's work current AI already does. That is the dimension the boxes above are about.
It cannot move the other three. Liability shield (18/20) is whether the law requires a licensed human to sign. Trust premium (7/20) is whether buyers specifically pay for a person. Judgment and accountability (15/20) is whether the role exists to own consequential calls. Those are facts about the occupation's standing, not about which tasks are in your week — a paralegal who does only trial exhibits still holds no licence. Together they are 40 of this occupation's 56 points (71%).
Embodiment (8/20) is also a property of the work rather than the worker, but we don't tag individual tasks as physical or not, so the picker can't tell you anything about it. That's a limit of this tool, not a claim.
Did we get the list right? Tell us what's missing — the tasks are written from the outside, and you're reading this from the inside.
Anesthesiologists SAFE
The moves above are yours to make. This is the other half: what would have to change in the world for the occupation itself to score higher. None of it is in any one person's gift, but it is where the floor actually comes from. Scores here are not a one-way ratchet. Only two of the five dimensions — task resistance and embodiment — track what machines can do. The other three track law, what buyers will pay for, and who is answerable, and those move in both directions, often in response to the same pressure AI creates. If every lever below landed, this occupation would score around 69/100 — SAFE.
Genuine two-tier structure: plain-film and screening-mammography volume is the commodity tier; the judgment tier is multiphase CT/MR, oncologic staging with prior comparison across years, and interventional/image-guided procedures (biopsies, drains, ablations, IR). If fellowship-trained IR and body/neuro subspecialty work becomes the dominant share of the job as commodity reads compress, measured task resistance rises without any new law. Watch subspecialty fellowship match data and the IR/DR residency split.
FDA continuing to clear all diagnostic imaging AI only as CADe/CADx adjuncts requiring a licensed interpreting physician's signature, plus CMS conditions of participation and state medical practice acts keeping the final report attributable to a board-certified radiologist. A concrete watchpoint: whether any AI vendor's product labeling ever permits autonomous reporting without physician review (as happened with IDx-DR in ophthalmology), and whether malpractice carriers (e.g. The Doctors Company, MedPro) write policies that price autonomous AI reads at all.
If the routine screening tier is largely pre-read by AI, the residual role concentrates in discrepancy resolution, indeterminate findings, tumor board participation, and signing off on AI outputs the radiologist did not generate — i.e. owning calls where the model and the clinical picture disagree. ACR's Assess-AI registry and 'AI QA' program formalizing radiologist responsibility for model drift and error adjudication would make this an explicit accountable duty rather than an informal one.
Same shift toward image-guided intervention — fluoroscopy, CT-guided biopsy, vascular access, ablation — is unpredictable-environment physical work. Only applies to the IR-heavy portion of the SOC code, not to teleradiology.
The limit. Trust premium has no realistic route: patients almost never choose their radiologist, rarely learn the name on the report, and the buyer is the hospital or payer optimizing cost per read — the same economics that made teleradiology outsourcing viable. The liability shield is already near ceiling and is the load-bearing dimension; if autonomous-read labeling and carrier coverage ever arrive, the whole 56 falls fast rather than degrading gradually.
| New York-Newark-Jersey City, NY-NJ | 1,950 | $465,780 +11% |
| Seattle-Tacoma-Bellevue, WA | 1,210 | $239,200 -43% |
| Dallas-Fort Worth-Arlington, TX | 770 | $156,800 -63% |
| Philadelphia-Camden-Wilmington, PA-NJ-DE-MD | 550 | $452,940 +8% |
| Houston-Pasadena-The Woodlands, TX | 500 | $131,470 -69% |
| Indianapolis-Carmel-Greenwood, IN | 400 | $239,200 -43% |
| Denver-Aurora-Centennial, CO | 340 | $344,270 -18% |
| Los Angeles-Long Beach-Anaheim, CA | 310 | $488,670 +16% |
| Minneapolis-St. Paul-Bloomington, MN-WI | 30 | $745,230 +77% |
| Sioux Falls, SD-MN | 50 | $669,020 +59% |
| Huntington-Ashland, WV-KY-OH | 60 | $592,050 +41% |
Futurism reports the CEO of New York City's public hospital system said it is prepared to begin using AI in place of radiologists.
Has AI actually changed your work? One tap, anonymous, and the running tally is public. Nothing else is asked of you.
Rather than check back: get the digest and we'll tell you what changed — or watch a single occupation from its own page.