Will AI Replace Radiologists? Shortage and Salary Data Say No
No, and the data goes the opposite direction from the fear. Radiologists are in short supply, residency positions are at record highs, and salaries are climbing, all while AI adoption in the field grows. AI is taking over first-pass image screening. It is not taking over the diagnosis.
This is the specialty most associated with "AI will replace X" headlines, dating back a decade to predictions that radiologists would be obsolete within five years. That prediction did not hold up. A 2025 workforce study found about 37,500 radiologists practicing in the US, with average salaries reaching $571,000, up 9% year over year (Fortune). A profession being automated away would not see pay move like this.
What the data actually says
The shortage is the load-bearing fact. Roughly half of all US radiologist job openings went unfilled in 2023 (Beckers). A 2025 workforce study projects the field will only reach about 47,000 practicing radiologists by 2055, while imaging demand is expected to grow by up to 27% over that same span, meaning the supply gap doesn't close, it widens (Fortune). Separately, projections point to a shortfall of 17,000 to 42,000 radiologists, pathologists, and psychiatrists combined by 2033 (Beckers).
Residency interest confirms the field isn't shrinking from the pipeline end either. In 2025, US diagnostic radiology residency programs offered a record 1,208 positions across all radiology specialties, up 4% from 2024, alongside all-time-high vacancy rates (Understanding AI). Medical students are choosing this specialty in growing numbers at the same time AI tools are becoming common in radiology departments.
Which tasks are exposed
First-pass image screening for common, well-defined findings is the clearest target: flagging likely-normal scans so a radiologist can move faster through a busy queue, triaging urgent cases (a suspected stroke, an active bleed) to the front of that queue, and catching some straightforward, textbook-pattern abnormalities before a human even opens the image.
AI adoption for this kind of task is real but uneven. One 2024 estimate put only 48% of radiologists using AI tools at all in their practice (Understanding AI), and a 2025 survey found that among radiology groups that have started piloting or deploying AI use cases, only 19% reported a "high" degree of success so far (Beckers). This is a field adopting the technology unevenly and cautiously, not one being swept aside by it.
Which tasks are protected, and why
Interpreting an ambiguous or atypical image, the kind that doesn't match a textbook pattern, stays with a radiologist, because that's precisely where current AI models are weakest and where the cost of a wrong call is highest. Integrating an image with the rest of a patient's clinical picture (their history, their symptoms, what the referring physician actually needs to know) requires context no imaging model has access to. And the final signed report that a referring physician acts on carries legal and clinical accountability that has to sit with a licensed person.
The clearest evidence that this protection is real, not just a professional talking point, comes from a 2025 multicenter diagnostic study: radiologists working with AI assistance statistically outperformed standalone AI on sensitivity for detecting intracranial hemorrhage, 98.91% versus 95.91% (PMC). AI alone is good. A radiologist plus AI is measurably better than either alone on one of the highest-stakes findings in the field.
What is already happening
AI is being used to cut workload on specific, repetitive screening tasks, with one estimate citing up to a 53% reduction in time spent on certain screening categories (Understanding AI). At the same time, the field is actively marketing AI tools as a partial answer to its own staffing shortage rather than as a headcount reducer, since radiology groups are the ones with unfilled positions, not a surplus of radiologists competing for scarce work (npj Health Systems).
Health systems facing that same shortage are increasingly framing agentic AI and workflow-automation tools as a way to keep imaging turnaround times reasonable given the staffing gap, not as a way to reduce the number of radiologists they employ (SATMED Health). That framing matches the salary and demand data above: a field cutting headcount does not typically see average pay climb 9% in a year.
What "AI-proof" actually means for this specialty
A growing field with unfilled positions doesn't mean every radiology task is safe, it means the specific tasks AI is good at (first-pass screening, triage, flagging likely-normal studies) are being absorbed inside a field with more total work than people to do it. That's a very different situation from an industry where AI is cutting into a fixed or shrinking amount of work. In the second kind of industry, task automation does translate more directly into fewer jobs. Radiology, imaging, and diagnostic support broadly sit in the first category, which is the specific reason the "radiologists are obsolete" prediction from a decade ago has aged so badly.
What to do about it
If you're a radiologist or in radiology training, the practical move is to get fluent with your department's AI triage and screening tools rather than treat them as a threat, because the data above shows radiologist-plus-AI outperforming either alone on the highest-stakes calls. Push for training if your group hasn't provided it, since the 2025 survey found only 19% of groups piloting AI reported high success, which points to implementation gaps more than tool limitations. And if screening volume is what's burning you out, that's the exact problem these tools are built to solve, given the shortage numbers above.
If you're earlier in your career and choosing a specialty, the residency numbers here (record positions, record vacancy rates) are a more reliable signal than a decade-old prediction about the field disappearing.
The tasks you keep decide how replaceable you are
The tasks you can still do without leaning on AI are what make you hard to replace here. The free 5-Day AI Reset is a five-email course built on that idea. On Day 2 you pick one task and do it without AI. Each day asks for one small change, and it stays useful no matter which way radiologists move.
Frequently asked questions
Will AI replace radiologists?
No. US radiology residency programs hit a record 1,208 positions in 2025, up 4% from 2024, alongside record vacancy rates (Understanding AI), and radiologists working with AI assistance still outperform standalone AI on key diagnostic measures (PMC).
Is there a radiologist shortage?
Yes. Roughly half of US radiologist job openings went unfilled in 2023 (Beckers), and projections point to a shortfall of 17,000 to 42,000 radiologists, pathologists, and psychiatrists combined by 2033.
How many radiologists actually use AI tools?
About 48% of radiologists reported using AI tools in their practice as of a 2024 estimate (Understanding AI), and among groups piloting AI, only 19% reported a high degree of success as of a 2025 survey, suggesting adoption is real but still maturing.
Does AI outperform radiologists on reading scans?
Not consistently on real clinical cases. A 2025 multicenter study found radiologists working with AI assistance had higher sensitivity for detecting intracranial hemorrhage (98.91%) than standalone AI alone (95.91%) (PMC), meaning the combination beats either one working alone.
Why did the "radiologists will be obsolete" prediction not come true?
Because it assumed image interpretation was a single task AI could fully absorb, when in practice most of a radiologist's value sits in integrating ambiguous findings with a patient's full clinical picture and taking legal accountability for the report, tasks current AI does not perform independently.
Want your own number instead of an industry average? Take the free How AI-Proof Is Your Job? assessment and see how your actual day-to-day tasks score.