Where a radiologist practises changes what a radiologist earns
$684,280top of the range in Colorado · middle $420,860 / yr
AI augments this role
Radiologists in the United States earn a median of $420,860 a year. Pay starts near $89,010. Pay reaches $684,280 at the top of the range in Colorado, the best-paying state for this work among those with at least 500 people in the job.
Source: U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025 (Radiologists, SOC 29-1224). Last checked 9 September 2026.
Entry level
$89,010
Top of the range · Colorado
$684,280
Education
Doctor of Medicine (MD) or Osteopathic Medicine (DO) + radiology residency
Wages — U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025 (Radiologists). Top of the range is the highest state-level figure among states with at least 500 people in the job. AI-impact rating is PayCrunch's editorial assessment. Updated September 2026.
🆕 New & Trending AI Tools for RadiologistReviewed September 2026
We track new AI-tool launches every week and refresh this list — here’s what’s gaining traction for Radiologist work right now.
AbridgeNEWEnterprise / see site
Ambient AI scribe that turns a patient conversation into structured clinical notes.
How a Radiologist uses it: document a visit automatically instead of charting after your shift
Microsoft Dragon CopilotNEWEnterprise / see site
Voice AI that dictates and drafts clinical documentation (successor to Nuance DAX).
How a Radiologist uses it: speak your notes and have the chart written and filed for you
Heidi HealthNEWFree / paid tiers
AI documentation tool built around clinician and nurse workflows.
How a Radiologist uses it: handle shift notes and handovers without manual write-ups
OpenEvidenceNEWFree for verified clinicians
AI that answers clinical questions from current medical evidence, with citations.
How a Radiologist uses it: check the latest evidence at the point of care in seconds
NotebookLMNEWFree / $7.99 mo
Google tool that answers questions grounded only in the documents you give it — with citations.
How a Radiologist uses it: load your own manuals, policies, or PDFs and ask questions that stay accurate to the source
SukiEnterprise / see site
AI voice assistant for clinical notes and coding.
How a Radiologist uses it: dictate notes hands-free and cut charting time sharply
NablaFree tier / see site
Ambient AI assistant that generates notes from the patient encounter.
How a Radiologist uses it: capture the visit and get a ready-to-review note in seconds
ChatGPTFree / $20 mo
The most-used AI assistant — writing, analysis, research, and images from a plain-language chat.
How a Radiologist uses it: draft emails and documents, summarize long files, and get instant answers to on-the-job questions
ClaudeFree / $20 mo
AI assistant known for careful writing, long-document analysis, and coding.
How a Radiologist uses it: analyze big reports or spreadsheets and turn messy notes into clean, finished writing
Radiologist: the Core is three remote days at about five and a half hours, and the impression is still yours
Radiologists on this page are SOC 29-1224. BLS OEWS May 2025 prints entry $89,010, a middle of $420,860, and $684,280 at the top of the range in Colorado. The radiologist chart’s top state is Colorado. The groups that pay toward that chip are paying for a reader who clears the list without signing a finding they did not see. Worklist software that floats a bleed to the top of the queue is useful. A generated impression that adds a mass you did not dictate is a miss in the other direction, and it is how a reader loses the shift that funds the top of the range.
The sitting in this section is the ABR Qualifying (Core) Exam in diagnostic radiology, not the Certifying Exam and not the oral exam that follows later. I read the ABR diagnostic radiology certification page and the July 2019 Core Exam policy PDF on 2026-09-19. The 2026 page reprints three remote days and approximately five and a half hours per day, scored pass or fail as a whole. The 2019 policy still reprints up to 12 hours administered over 1.5 days, a minimum of 60 scorable units per category except physics at least 90, and at least 630 total units including independent RISE. Those clocks are not averaged. Do not turn 5.5 times 3 into the 2019 twelve-hour day, and do not turn the 2019 day into the 2026 remote format. The 2026 page I used did not print a Core item count. Third-party “615 questions” or “15 hours 44 minutes” clocks stay off this page. Certifying is a different sitting: approximately seven and a half hours, one day, four modules. The oral exam described for after 2027 is seven 25-minute sessions. It is not this Core. Physics is embedded. A generated “separate physics day you can condition” is not what these pages say. No 70 percent cut was printed. Do not invent one.
Core clocks that are not averaged
2026 format on the page I read3 remote days · about 5.5 hours each
2026 scoringPass/fail as a whole
2019 policy, printed beside itUp to 12 hours · 1.5 days · ≥630 units
2019 category floor≥60 units · physics ≥90
Certifying, different dayAbout 7.5 hours · four modules
ABR pointed readers to the Core Exam Guide for the item detail the 2026 overview does not print. A vendor pie chart of organ-system percents is not that guide. Do not study the Certifying modules as if they were Core days.
Triage the list. Dictate the finding. Edit the impression.
Aidoc, Viz.ai, and RapidAI, where the hospital has deployed them, reorder the list. A suspected large-vessel occlusion or a bleed rises. You still open the study. The algorithm did not sign the report. PowerScribe, or the reporting system the group already uses, will offer an impression from what you dictated. The earning habit at $420,860 and above is a fixed edit pass: every sentence in the impression must point at a finding you said, and every critical finding you said must appear in the impression. The model is allowed to show you the mismatch. It is not allowed to add a nodule, a fracture, or a recommendation for biopsy.
Prompt — impression against the words you dictated
Here is the indication I typed and the findings I dictated. No images and no patient name. List findings in the dictation that are missing from the draft impression, and list any impression sentence that is not in the dictation. Do not add a finding. Do not recommend a biopsy, a follow-up interval, or a comparison I did not state. If I pasted an accession number, stop.
That prompt runs inside the reporting tenant, or not at all. DICOM does not go to a consumer chatbot. A photo of a screen is still the study. Comparison dates come from the prior report you opened, not from a model’s guess that “lungs are unchanged.” Unchanged is a statement you make after you looked. Groups paying toward Colorado’s $684,280 chip measure turnaround on the studies that matter and discrepancy rates on the ones you signed. A reader who is fast because the impression writes itself, and wrong because they did not read it, will not keep that shift. A reader who uses triage to open the right study first, and who deletes the extra sentence, will.
Peer review and the study day are different files
Discrepancy logs are an earnings tool if you actually trend them. Export a month of your addenda with the accession stripped and the reason coded by you: missed finding, wording, comparison not opened, recommendation you did not mean. Ask Claude, in the group’s approved workspace, to count by reason. You coded the reason. The model does not recode a miss as “wording” because that feels kinder. That count is what you take to the section chief when you want the overnight list or the procedural time that pays above the middle chip. It is also what you take to yourself. A log you will not show anyone is not a log.
Prompt — Core facts, no invented item count
Using only the ABR pages I uploaded, state the 2026 Core format and the 2019 unit floors as separate lines. Do not average 5.5 hours times 3 with 12 hours. Do not invent an item count. Do not write a 70 percent cut. Do not describe the Certifying Exam as if it were the Core. If the upload does not contain those pages, stop.
NotebookLM is for that study file, not for a teaching-file case that still has a medical-record number burned into the image. Strip the header before a case enters any deck, including a deck that never leaves your laptop. Physics at the 2019 floor of at least 90 units is embedded in the Core. It is not a day you can fail and still pass the rest, because the 2026 scoring I read is pass or fail as a whole. A study bot that offers “conditioned physics” is describing a different era or a different board. Certifying’s four modules and the later orals are real, and they are later. Spending Core week on a seven-and-a-half-hour certifying outline is how you waste one of the three remote days.
Critical-result callbacks are a line in the report, not a memory. The time, the person you reached, and the finding you spoke are typed by you after the call. A scribe template that inserts “results discussed with the referring physician” without those three facts is a line you delete. Groups that staff the shifts near Colorado’s chip audit that line. A callback you cannot reconstruct from the chart is a callback you did not document, and the worklist tool that floated the study does not make the phone call.
A shift that protects the wage has a reordered list and a signed impression you can defend. Triage floated two studies. You read both, and you read the ordinary studies the algorithm did not love. The impression matches the dictation. The critical result was called by you, and the time of the call is in the report because you typed it. No screen photo left the reading room. The Core, when you are still in it, is three remote days at about five and a half hours, pass or fail as a whole, with the 2019 unit floors sitting beside that format and not blended into it. Colorado’s $684,280 figure is the top of the range this page already prints. An algorithm subscription does not add a digit to it.
Wage chips already on this page
Entry$89,010
Middle$420,860
Top of the range · Colorado$684,280
BLS OEWS May 2025, Radiologists (SOC 29-1224). Colorado leads the radiologist chart. Exam fees are not on this card.
Core figures are from the ABR diagnostic radiology page and the July 2019 policy PDF, read 2026-09-19. Three remote days at about 5.5 hours, pass/fail as a whole. The 2019 floors are printed beside that format, not averaged with it. No item count was invented. No ASIN. Ads unchanged.
The top of Radiologist pay — and how to get there with AI
$684,280what Radiologist pay reaches in Colorado
Highest state-level top-of-range annual wage for Radiologists, among states with at least 500 people in the job. U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025.
And the role it leads to — Cardiologists — reaches $934,460 in Georgia.
$89,010entry$420,860middle$684,280ceiling
Two radiologists reading the same studies are paid very differently depending on whether they sit in a salaried hospital post, an independent practice with an ownership track, or a subspecialty service that also performs procedures.
Reading volume is the wrong thing to optimise. Interpretive reports of imaging findings are being drafted from structured templates and dictation faster every year, and triage software already sorts the worklist, so a career built purely on throughput is competing against a falling price. What holds value is the part that requires a physician to be accountable: performing procedures rather than only interpreting them, recognising and treating complications during and after them, being the person a referring clinician telephones about an ambiguous study, and coordinating radiological services with the rest of a hospital's activity. Settings differ enormously in how much of that work they let you own.
Your playbook, by where you are now
Just startingChoose the fellowship for the practice you want
Pick a subspecialty by the setting it leads to — interventional and neuro open procedural and academic doors that a general reading post does not.
Build a report style referring clinicians act on without a phone call: findings, then a clear impression that answers the question actually asked.
Practise obtaining histories properly from electronic records and by contacting the referring clinician, because an unexplained study is the main cause of a hedged report.
Learn the picture archiving and communications system deeply, including how images and reports move to outside institutions.
Use Claude or Microsoft Copilot to pressure-test a draft impression against the differential you considered, then commit to your own wording.
What proves it: A subspecialty fellowship and a referrer base who ask for your reads by name.
Realistic span: training and the first two years after
A few years inCompare offers on structure, not headline pay
Ask every prospective employer three questions: is there a partnership or ownership path, who bills for procedures, and who controls the schedule.
Take the procedural work rather than routing it away, since procedures are where a radiologist's revenue and clinical standing are hardest to substitute.
Track your own case mix and turnaround in Microsoft Excel, because an independent practice negotiation runs on your numbers, not the group's.
Instruct radiologic staff in the techniques, positions and projections you need; a technologist trained by you produces studies you can read fast and confidently.
Join the quality improvement work where high-risk error areas are discussed, which is the committee that decides how services are structured.
What proves it: An offer or contract with a defined ownership path and procedural scope in writing.
Realistic span: years three through eight after fellowship
ExperiencedOwn a service line
Take a service-line or section leadership role and coordinate radiological services with surgery, oncology and emergency care directly.
Negotiate the group's contracts and equipment plan, since the physician who shapes what the department buys shapes what it can bill.
Build a subspecialty referral practice that draws work in from outside the institution rather than waiting for the internal worklist.
Set the reporting standards and structured templates the whole section uses, working from Epic Systems and the digital image processing tools already in place.
Weigh geography honestly: Colorado leads the states for radiologist pay, and interventional cardiology is the adjacent field for those who prefer being in the suite full time.
What proves it: Section or service-line leadership with contract and equipment decisions in your hands.
Realistic span: nine years in and beyond
The next 90 days
Over the next ninety days, take your own practice apart on paper. Split your work into three columns: studies anyone with your training could read, studies that needed your subspecialty, and procedures or consultations where you were personally accountable for the outcome. Note who bills for each column and who decides your daily volume. Then talk to two radiologists in different settings from yours — an independent group and a hospital department — and ask them the same three questions about ownership, procedural billing and schedule control. Most radiologists have never made this comparison explicitly, and it is the comparison that separates the middle of this range from the top far more reliably than reading faster does.
Wage figures: BLS OEWS, May 2025. The playbook is PayCrunch editorial guidance, not a guarantee of pay or placement.
Every figure is the national median from the U.S. Bureau of Labor Statistics (OEWS) shown on that role’s own page.
Never used AI before? Start here (2 minutes).
Start with the AI your department already licenses. Most radiology groups now run AI reporting (Rad AI, Nuance PowerScribe) and triage (Aidoc). Turn on the AI-generated impression on one shift and treat it as a draft you verify and edit — you stay in full control of the signed report.
For learning and general reference (never patient data), open OpenEvidence or UpToDate, or ChatGPT for a plain-language refresher on a guideline. Keep everything with patient identifiers inside your approved clinical systems. AI is the resident who preps the report; you are the attending who signs it.
The one rule, forever: AI findings are decision support only — the interpreting physician must independently review every image and remains legally responsible for the signed report. Use only HIPAA-compliant, FDA-cleared tools inside your PACS; never paste protected health information into a consumer AI tool, and never let a 'normal' AI result stop you from looking.
The plays — exact steps, exact prompts
Do these in order. Each one is copy-paste ready. You do not need to know anything about AI going in.
1
Draft reports at dictation speed with AI reporting
Why this pays: Radiology comp is largely RVU-driven — more accurate reads per session means more income. AI report generation that produces a solid draft impression is the single biggest throughput lever, pushing toward the top of the range.
Rad AI OmniNuance PowerScribeRADPAIR
1
Enable AI-generated impressions and structured reporting in Rad AI Omni or Nuance PowerScribe. Read the images first, then let the tool draft the report language for you to verify and sign.
2
Standardize your high-volume study types with clean templates so the AI drafts consistently.
Copy-paste this prompt
Draft a structured reporting template for [CT chest with contrast] that follows [Fleischner Society] recommendations for incidental pulmonary nodules. Include section headings, normal-template language, and an impression section with follow-up wording keyed to nodule size. General template only, no patient data.
Build reusable templates from general guidelines. Never put a real patient's data into a consumer tool.
What you'll haveAccurate reports finalized faster on every study — the throughput that drives RVU-based income toward $594,410.
2
Add an AI triage and detection safety net
Why this pays: Fewer misses means fewer malpractice events and faster STAT turnaround — both protect and grow high-decile income. AI worklist triage surfaces the critical study first so nothing waits.
AidocAnnalise.aiViz.ai
1
Let Aidoc or Viz.ai prioritize your worklist so intracranial hemorrhage, PE, or large-vessel occlusion jumps the queue and gets read first.
2
Treat detection AI (e.g., Annalise.ai) as a second reader that flags findings — you confirm or overrule every one. Use it to catch, never to skip.
3
Audit the AI's flags against your reads periodically so you know its real-world sensitivity and where it's weak on your equipment.
What you'll haveA safety net that speeds critical results and reduces misses — protecting both patients and your income.
3
Subspecialize and let AI clear the routine
Why this pays: Subspecialty reads — neuro, breast, cardiac, IR — pay more than general work. Using AI to speed the routine, normal studies frees your time and attention for the complex, higher-value ones.
Gleamer (BoneView)Qure.aiOpenEvidence
1
Deploy narrow, high-confidence tools like Gleamer BoneView for fractures or Qure.ai for chest X-ray to move quickly through routine normals — always confirming.
2
Build deep expertise in a subspecialty and let AI handle the volume that doesn't need it.
Copy-paste this prompt
Act as a radiology educator. Build me a 90-day self-study plan to sharpen my [neuroradiology stroke imaging] reads: key patterns, common misses, must-know guidelines, and 5 landmark papers to read. General education only, no patient data.
Depth in a subspecialty is durable value AI can't replicate. Keep all study material patient-free.
What you'll haveMore time and focus on high-value subspecialty reads — the mix that lifts you to the top of the pay band.
4
Query the latest evidence at the workstation
Why this pays: Better recommendations and appropriate follow-up build referral reputation and reduce liability — the reputational engine behind a busy, well-paid practice.
OpenEvidenceUpToDateChatGPT
1
When a management or follow-up question comes up, check current guidance fast — in general terms only.
Copy-paste this prompt
For an incidental [3 cm adrenal nodule] found on CT, summarize the current recommended characterization workup and follow-up imaging guidance, including the thresholds that change management. Cite the guideline source. General guidance only, not a specific patient.
Ask in general terms; never enter identifiable patient details into a consumer AI tool.
2
Fold the guideline-concordant recommendation into your report. Consistent, evidence-based follow-up advice earns referrer trust.
What you'll haveGuideline-concordant recommendations delivered instantly — sharper reports and a stronger referral reputation.
5
Moonlight teleradiology efficiently
Why this pays: AI-augmented reporting makes each read faster, which makes remote night/weekend teleradiology worthwhile. Extra reads on top of a day practice are a direct path into the $594,410 tier.
Rad AI OmniNuance PowerScribeteleradiology PACS
1
Use the same AI reporting tools remotely to keep per-study time low so moonlighting reads are actually worth your evening.
2
Vet any teleradiology contract carefully before you sign.
Copy-paste this prompt
List the questions I should ask before joining a teleradiology group: RVU or per-study rates, malpractice tail coverage, state licensing and credentialing support, which PACS and AI tools are provided, expected volume, and turnaround-time SLAs.
Efficiency only pays if the contract terms and malpractice coverage are right. Do the diligence.
What you'll haveSustainable extra reads at a fair rate — the incremental income that pushes total comp into the top of the range.
6
Lead AI adoption and quality assurance for your group
Why this pays: The radiologist who validates tools, builds templates, and monitors performance becomes indispensable — a route to partnership, leadership, and the group-wide efficiency that lifts everyone's earnings.
ACR Assess-AI / ACR DSIAidocRad AI Omni
1
Propose a rigorous validation before any new tool goes live.
Copy-paste this prompt
Draft a 1-page proposal to validate a new AI [pulmonary embolism triage] tool before clinical use in our group: what metrics to measure (sensitivity, specificity, turnaround time), how to run a shadow-mode pilot on our own cases, and how to monitor performance after go-live.
Validate on your own patient population and scanners — vendor numbers rarely transfer exactly.
2
Own the report templates and the AI QA process for the group. Leadership plus efficiency is what earns partnership-level comp.
What you'll haveA validated, monitored AI program you lead — raising the whole group's throughput and your path to partnership.
Your 12-month sequence to the top of the range
How the plays above stack into a path from median pay toward the $594,410 tier.
Month 1
Turn on AI report drafting on your highest-volume study types. Verify and sign every one; measure time saved.
Months 2-3
Integrate AI worklist triage and treat detection AI as a confirming second reader.
Months 3-6
Deepen a subspecialty; use narrow AI tools to clear routine normals and focus on complex reads.
Months 6-9
Add point-of-care evidence lookups to sharpen recommendations and referral reputation.
Months 9-12
Consider AI-efficient teleradiology moonlighting on well-vetted contracts for incremental RVUs.
Year 2
Lead your group's AI validation and QA — the leadership route to partnership and top-of-range comp.
What Radiologists earn by state
These are the Bureau of Labor Statistics’ own figures for Radiologists, state by state — not a cost-of-living adjustment applied to the national number. Only states employing at least 500 people in the occupation are shown, because a state median drawn from a handful of workers is noise rather than a signal.
Pennsylvania
$455,380
highest of them · +8% vs the national median
Ohio
$260,000
lowest of the 9 states that qualify · -38% vs the national median
The same job pays $195,380 more a year at the median in Pennsylvania than in Ohio — 75% higher. That gap is what the Bureau measured, before any question of what it costs to live in either place. The top-of-range figure quoted at the head of this page, $684,280, is a different statistic in a different place: it is the 90th-percentile wage in Colorado. The state that pays the typical worker most and the state where the best-paid go highest are not always the same one.
Source: U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025, SOC 29-1224. 9 states clear the 500-employee reporting floor for this occupation; those below it are left out rather than shown with a wide error band.
Free data. Use any of it.
PayCrunch publishes verified, BLS-sourced salary + AI-playbook data on 1,000+ professions — free, no signup.
No — and the field's own data bears this out: demand for radiologists is strong even as AI adoption accelerates. AI reads pixels; radiologists integrate images with the whole clinical picture, perform procedures, communicate with referrers, and carry legal responsibility for every diagnosis. AI is augmentation. The radiologists who use it read more efficiently and safely; those who ignore it fall behind on throughput.
Can I trust an AI-drafted report or a 'normal' AI result?
Only as a draft and a prompt, never as the final word. You must review every image yourself and verify every line of an AI-generated report before signing — the liability is entirely yours. A 'normal' AI result must never stop you from looking; these tools miss findings, and their performance varies by scanner and population.
Is it safe to use ChatGPT in radiology?
Not with any protected health information. Consumer AI has no place touching identifiable patient data. Use FDA-cleared, HIPAA-compliant tools inside your PACS for clinical work, and reserve general tools like ChatGPT, OpenEvidence, or UpToDate for education and guideline lookups phrased in general terms.
How does AI actually increase a radiologist's pay?
Comp is largely RVU-based, so throughput matters. AI report generation cuts dictation and editing time, and AI triage front-loads the critical studies — letting you read more studies accurately per session and safely take on subspecialty or moonlighting volume. It's efficiency, not shortcuts, that moves income toward the top of the range.
Which AI tool should a radiologist prioritize?
Whatever your group already runs — most start with AI reporting (Rad AI Omni or Nuance PowerScribe) because it saves time on every single study, then add triage (Aidoc, Viz.ai) for critical findings. Learn the reporting tool first; it has the broadest daily impact.
Methodology & sources
Salary (median, 10th, top of the range) — U.S. Bureau of Labor Statistics, OEWS.
By state — the Bureau of Labor Statistics’ own state medians, limited to states employing at least 500 people in the occupation. No cost-of-living arithmetic is applied to a wage anywhere on this page.
The plays — PayCrunch's own step-by-step guidance using publicly available AI tools. Tool names/URLs are real and current as of August 2026; prompts written to work as-is. Verify any professional output before relying on it.