The FDA has cleared more than 700 AI-enabled medical devices through 2024. Almost none replace clinicians. The first wave of medical AI coverage promised computers would put radiologists out of work. They did not. The second wave is doing something more useful and less photogenic — becoming infrastructure inside diagnostic workflows that most patients never see.
Where AI is already routine
Several categories of AI tools are FDA-cleared and in everyday clinical use:
Diabetic retinopathy screening
IDx-DR was the first autonomous AI diagnostic system the FDA cleared, in 2018. It reads retinal images in primary care settings, no ophthalmologist required. Patients get a screening result during a routine GP visit instead of waiting weeks for a specialist referral.
Stroke triage
Tools like Viz.ai analyze head CT scans, flag suspected large-vessel occlusions, and alert stroke teams in minutes. Time-to-treatment is the single biggest predictor of stroke outcomes — minutes saved on triage translate into measurable functional recovery.
Chest X-ray analysis
Several cleared systems triage chest X-rays for pneumothorax, nodules, and other findings. They do not replace the radiologist’s read. They reshuffle the queue so urgent findings get seen first.
Mammography assistance
AI-aided mammography reading is now routine across much of Europe. US rollout has been slower, partly because of reimbursement uncertainty rather than evidence.
Cardiac imaging
AI tools measure ejection fraction and valve function automatically from echocardiograms — work that used to take a sonographer significant time.
The shift to ambient documentation
The most disruptive use of AI in medicine right now is not diagnostic. It is clerical. Ambient documentation tools listen to patient visits, generate draft notes, and free clinicians from typing. Major US health systems have rolled these tools out broadly. Reduced clinician burnout is the headline benefit; whether they affect diagnostic quality is being actively studied.
Where LLMs are landing
Large language models have moved into specific clinical workflows:
- Differential diagnosis suggestions for complex presentations
- Summarization of long patient histories before a visit
- Drafting patient-facing communications — discharge instructions, MyChart replies
- Drug interaction checks during prescribing
Some studies show LLMs matching or beating physicians on USMLE-style exams. Generalizing from exam performance to actual clinical care is more complicated. Patients are messier than test questions.
Where AI is still unreliable
- Rare conditions underrepresented in training data
- Populations underrepresented in training data — multiple studies have found AI dermatology tools perform worse on darker skin
- Situations requiring physical examination — a chatbot cannot palpate an abdomen
- Any case where the input data is poor — bad image, incomplete history
The regulatory picture
The FDA has cleared hundreds of AI-enabled devices, though most are narrow tools rather than autonomous decision-makers. The EU’s AI Act places most medical AI in the high-risk category, requiring conformity assessment, documentation, and human oversight. Practical effect: developers building for global markets are designing to the stricter standard.
What this means for patients
Most patients have probably already had AI involved in their care — they did not know. If your chest X-ray went through a queue that prioritized critical findings, if your last mammogram benefited from a second AI read, if your retinal screen happened in primary care — you have encountered medical AI. The benefits are mostly invisible: faster identification of urgent cases, less missed pathology, less burnt-out staff.
What patients should still expect
A human clinician interpreting your case, making decisions, and answering your questions. AI in current use is a tool clinicians use. It is not a replacement for them. If a clinician ever tells you a diagnosis came purely from an AI output with no human review, ask questions.
The bottom line
The interesting AI-in-medicine story now is logistics, triage, and workflow — quieter than the original radiology-replacement narrative, with broader impact on what care actually feels like.
