"Will AI replace radiologists?" is one of the most searched questions in medicine, and it is the first thing many students ask before choosing radiology. Here is what the data actually say in 2026, what AI does well and badly today, and what every resident should learn now.
AI in radiology by the numbers
| Fact | Figure |
|---|---|
| AI/ML-enabled devices authorised by the US FDA, 1995 to end of 2025 | 1,430 |
| Of these, devices in radiology | 1,094 (about 76%) |
| AI devices authorised in 2025 alone | 331 |
| Radiology's rank for average US physician pay (Medscape 2026 report) | 3rd, about US$571,000 a year |
Radiology has dominated medical AI from the start because images are digital, plentiful and already labelled by reports. Yet a decade after the famous 2016 prediction that we should "stop training radiologists", demand for radiologists and their pay have kept rising.
What AI does in radiology today
| Use | Examples |
|---|---|
| Triage and worklist prioritisation | Flagging suspected intracranial haemorrhage, large vessel occlusion, pulmonary embolism or pneumothorax so they are read first |
| Detection ("second reader") | Lung nodules on chest CT and radiographs, fractures, breast cancer on mammography |
| Quantification | Brain volumes, cardiac function, liver fat, bone age, coronary calcium |
| Image reconstruction | Deep-learning reconstruction that lowers CT dose or shortens MRI scan time |
| Workflow and reporting | Protocoling, hanging protocols, structured reports and AI-drafted report impressions |
Where AI still struggles
- Narrow tasks: most products answer one question (is there a bleed?), while a radiologist reads the whole study and the whole patient.
- Generalisation: performance can drop on scanners, populations and protocols different from the training data.
- False positives and automation bias: people tend to over-trust confident software; you remain responsible for the report.
- Evidence gaps: many cleared tools have limited published outcome data, and clearance is not the same as proof of patient benefit.
- Integration and cost: tools must fit the PACS and RIS, and someone must pay and monitor them.
So will AI replace radiologists?
Not in the foreseeable future, but it will change the job. The likeliest picture is radiologists who use AI replacing radiologists who don't. AI is taking over repetitive measurements and the first look at simple, high-volume findings, which frees time for complex cases, procedures, multidisciplinary meetings and talking to clinicians and patients. Interventional radiology, ultrasound and problem-solving imaging are especially hard to automate.
What residents should learn now
- Statistics that judge AI: sensitivity, specificity, positive and negative predictive value, ROC curves and AUC, and how prevalence changes PPV.
- How models fail: dataset bias, shortcut learning and drift over time.
- Using tools safely: read the image first, then look at the AI result; document when you disagree.
- Basic informatics: DICOM, PACS, structured reporting and how data are anonymised.
- Your core skills: exams still test image interpretation, not software. Strong fundamentals make you the person who can supervise the machine.
Statistics questions appear in FRCR 2A and FCPS papers; our question bank includes a statistics and contrast media section.
Frequently asked questions
Will AI replace radiologists?
Current evidence says no. AI performs narrow tasks such as flagging a bleed or a nodule, while radiologists interpret whole studies in clinical context, perform procedures and take responsibility. AI is changing the work, and radiologists who use it well will be in demand.
How many AI tools are approved for radiology?
A 2026 analysis found 1,430 AI/ML-enabled devices authorised by the US FDA up to the end of 2025, of which 1,094 (about 76%) were in radiology.
Is radiology still a good career choice with AI?
Yes. Imaging volumes keep growing, and radiology ranked third for average US physician compensation in Medscape's 2026 report. Choose it if you enjoy imaging, problem solving and procedures, and plan to learn AI as a tool.
Do I need to learn coding to use AI in radiology?
No. Most radiologists need to understand how to evaluate and use AI tools safely, not how to build them. Coding helps if you want to do research or informatics.
Sources: AuntMinnie, Radiology dominates thirty years of FDA AI device approvals (July 2026, reporting a Cureus study); AuntMinnie, Radiology among top specialties for pay (Medscape 2026 report).