Question 7 of 20 · Careers and specialties
Diagnostic specialties under pressure
What happens to radiology, pathology and dermatology?
2 min read · Well measured today
These fields contain highly digitized tasks and a lot of AI activity, and they vary enormously inside. They are under real pressure. They are not obsolete, and we will not call them either safe or finished.
What we can measure today
- Established now
Radiology accounts for 1,039 of the FDA-authorized AI devices counted through 2025, far more than any other specialty.
What this does not prove: Device counts show where products are being built, not how many radiologists are needed.
Stanford HAI, 2026
The 2026 AI Index Report, Chapter 6: Medicine
Scope: Global research, with U.S. data on FDA authorizations.
How it was done: An annual review that summarizes studies by other researchers.
Limits: A summary of studies with different designs. Each headline figure comes from one setup and does not generalize on its own.
Checked against the original on November 6, 2026.
Open the source (opens in a new tab) - Established now
BLS May 2025 median pay is $420,860 for radiologists and $312,400 for pathologists, and it projects radiologist employment to grow 3% from 2025 to 2035.
What this does not prove: Today's snapshot and a projection. Neither is evidence about pay in 2036.
U.S. Bureau of Labor Statistics, Last modified August 27, 2026
Occupational Outlook Handbook: Physicians and Surgeons
Scope: United States.
How it was done: Federal employment and wage statistics. Pay is the May 2025 median, half earn more and half earn less. Growth is a 2025 to 2035 projection.
Limits: A median across all ages, regions and employers, before tax. A projection, not a promise. It tells you nothing about one person's income. The outlook text on this page names aging and chronic illness as demand drivers, and does not mention AI.
Checked against the original on November 6, 2026.
Open the source (opens in a new tab) - Established now
Stanford notes that FDA clearance does not equal clinical adoption, since financial, operational and institutional barriers stand in between.
Stanford HAI, 2026
The 2026 AI Index Report, Chapter 6: Medicine
Scope: Global research, with U.S. data on FDA authorizations.
How it was done: An annual review that summarizes studies by other researchers.
Limits: A summary of studies with different designs. Each headline figure comes from one setup and does not generalize on its own.
Checked against the original on November 6, 2026.
Open the source (opens in a new tab) - Established now
FDA authorized DermaSensor in 2024 to help non-dermatologist physicians decide on referral for suspicious skin lesions. It caught 95.5% of cancers but correctly cleared only 20.7% of benign lesions.
What this does not prove: Not a screening tool or a sole basis for a decision, and with less data on darker skin.
U.S. Food and Drug Administration, January 12, 2024
De Novo DEN230008: DermaSensor
Scope: Physicians who are not dermatologists, patients 40 and older, lesions already judged suspicious. Study sites in the U.S. and Australia.
How it was done: FDA's order and review, with the DERM-SUCCESS study of 224 malignant and 1,355 benign lesions.
Limits: Not a screening tool and not a sole basis for a decision. Less melanoma data on darker skin, and the study was manufacturer funded.
Checked against the original on November 6, 2026.
Open the source (opens in a new tab) - Established now
In a Swedish randomized trial of 105,934 women, AI-supported reading cut mammogram screen readings by 44% and found 6.4 cancers per 1,000 against 5.0. Interval cancers were non-inferior, at 1.55 against 1.76 per 1,000.
What this does not prove: Radiologists still read every exam. Interval cancers are not proven lower, and there's no mortality data. This shows reading work can shrink. It does not show what happens to jobs.
The Lancet, and Lancet Digital Health, January 31, 2026, with results from March 2025
MASAI randomized trial of AI-supported mammography screening
Scope: 105,934 women randomized at four screening sites in southwest Sweden.
How it was done: Randomized trial. AI triaged exams to single or double reading and flagged findings. Radiologists still read every exam. Read from the published abstracts.
Limits: One vendor and one region of one country. Interval cancers were non-inferior, not proven lower. No mortality data in these reports.
Checked against the original on November 6, 2026.
Open the source (opens in a new tab) - Unknown
We have not yet verified a rigorous study of what AI has done to radiologist employment.
What may change
- Routine reads, second opinions, triage, reporting and throughput.
What may endure
- Verification, interventions, complex cases, safety and clinical integration, although these may also automate.
The hard questions
- Would I still love the specialty if most of its routine work changed?
- Where would the expertise move?
Evidence worth watching
Prospective trials · Payment rules · Productivity · Error rates · Training demand
Sources last checked November 6, 2026
Next review by February 6, 2027.
What changed, and how we decideCareer guidance, not medical, legal or financial advice. For a major loan or licensing decision, talk with a qualified professional, and bring your questions from here.
Do you disagree?
If you have counterevidence, or we got something wrong, we want to see it. Don't take our word for any of this.
Send us the evidence