AI in Radiology: What Patients Should Know About AI-Read Scans

You get a letter after your mammogram, or a portal message after a CT scan, and somewhere in the fine print it mentions “computer-aided detection” or “AI-assisted reading.” It is natural to wonder: did a computer read my scan instead of a doctor? And if it got something wrong, who is responsible?

The short version is reassuring. In the US today, AI in radiology works mostly as an assistant. It helps a radiologist spot, sort and measure things faster, but a licensed physician still makes the call and signs your report. This guide explains what the tools do, what the best research shows so far, and the questions worth asking at your next scan.

This is general information, not medical advice. Talk to your doctor about your own results.

What AI actually does with your scan

Radiology is the medical specialty that has adopted AI the most. Analyses of the FDA’s public list of AI-enabled medical devices consistently find that radiology accounts for roughly three-quarters of all AI device authorizations. Those tools fall into a few broad jobs:

• Detection (a second set of eyes). The software marks areas on a mammogram, chest X-ray or CT that look suspicious, such as a possible nodule or calcification, so the radiologist looks twice.
• Triage (sorting the queue). Some tools scan incoming CT images for urgent findings like a possible brain bleed or blood clot in the lungs and move those cases to the top of the worklist so they get read sooner.
• Measurement. AI can measure a lung nodule, estimate breast density or track a tumor’s size between scans, work that is slow and tedious by hand.
• Image quality. Some tools reduce noise in images, which can help produce clear pictures with shorter scan times or lower doses, depending on the device.
• Report support. Newer tools draft parts of the written report or suggest follow-up recommendations, which the radiologist then edits.

Most of these tools are built with deep learning, a type of machine learning trained on large sets of labeled images. If you want the background, see our explainer on machine learning vs deep learning.

Who reads your scan and who is accountable

AI tools on the market in the US are generally cleared by the FDA as aids to a physician, not as replacements. The radiologist reviews the images, decides whether to accept or ignore the AI’s marks and signs the final report. The American College of Radiology (ACR) has long taken the position that radiologists remain responsible for patient care as they adopt AI.

In practice that means:

• The signed report is the radiologist’s. If something is missed or misread, the physician and the practice remain accountable under the usual standards of care.
• The AI can be wrong in both directions. It can flag something harmless (a false positive) or miss something real (a false negative). The radiologist’s job is to catch both.
• The facility chooses and monitors the tool. Hospitals and imaging centers decide which products to buy and should check how they perform on their own patients and equipment.

Key takeaway: In US imaging today, AI suggests and a radiologist decides. The signed report is the physician’s responsibility, and the facility is responsible for choosing and monitoring the AI it uses.

AI in mammograms: what the research shows

Breast screening is where the strongest evidence exists, thanks to large trials in Europe, where screening mammograms are typically read by 2 radiologists.

The MASAI trial in Sweden randomly assigned more than 100,000 women to either AI-supported screening or standard double reading by 2 radiologists. Key published findings:

• Interim safety results (The Lancet Oncology, 2023) found AI-supported screening was safe, detected more cancers and cut radiologists’ screen-reading workload by about 44%.
• Final results (The Lancet, 2026), as reported by Lund University, showed 29% more cancers detected and 12% fewer interval cancers (cancers found between scheduled screenings), without an increase in false positives.

Those results are encouraging, but keep 3 caveats in mind. The trial took place in a Swedish screening program with its own workflow, it tested a specific commercial product, and US screening is usually read by 1 radiologist rather than 2. Results may differ in other settings.

Separately, a federal rule now gives every US mammography patient more information. Under FDA’s updated Mammography Quality Standards Act (MQSA) regulations, enforced since September 10, 2024, your lay summary must tell you whether your breast tissue is dense or not dense. Dense tissue can hide cancers on a mammogram, and some AI tools estimate density, but the requirement applies whether or not AI is used.

AI in CT scans and X-rays

Outside mammography, AI shows up most often in:

Triage tools change the order of the queue, not the diagnosis. If your scan is not flagged, it is still read by a radiologist in the normal order. It just was not pushed to the front.

The real risks patients should understand

AI in radiology is not risk-free. The main issues researchers and regulators watch are:

• Performance gaps across groups. A tool trained mostly on images from 1 population or 1 brand of scanner may perform worse on others. Ask whether the facility has checked results on its own patients.
• Over-reliance. A busy reader might trust the software’s “all clear” too much. Good practices train radiologists to treat AI as advice.
• More follow-up tests. Highly sensitive tools can find tiny findings that lead to extra scans or biopsies, some of which turn out to be nothing.
• Drift over time. Software that worked well at launch can slip when scanners, protocols or patient mix change. Facilities should monitor performance continuously.

None of this means you should refuse an AI-assisted read. It means the human review step matters, and so does the facility’s quality program. For the bigger picture of where AI helps and struggles in medicine, see AI in healthcare in 2026.

Your rights: consent, disclosure and access

Patients often ask whether they have to be told when AI is used. Today:

• There is no single federal rule requiring imaging centers to tell you that AI assisted with your read. Practices vary, and some states have passed laws on AI disclosure in health care, mostly aimed at generative AI that communicates with patients. Our guide to US state AI laws tracks the trend. This is not legal advice.
• You have a right to your records. Under HIPAA, you can request copies of your imaging reports and, in most cases, the images themselves.
• Results may reach you fast. Federal information-blocking rules mean many results appear in your patient portal quickly, sometimes before your doctor has called you. If a report uses unfamiliar terms, wait for your clinician or message them rather than relying on internet searches. Our guide on asking an AI chatbot about symptoms covers how to use AI to prepare questions safely.

Questions to ask at your next scan

You do not need to be technical to ask good questions. Try these:

1. Do you use AI software to help read this type of scan?
2. Is it FDA-cleared, and what does it do: detect, triage or measure?
3. Does a radiologist review every image, even if the AI finds nothing?
4. How do you check that the software works well for patients like me?
5. If the AI flags something the radiologist disagrees with, what happens?
6. Who should I contact if I have questions about my report?

A good imaging center will answer clearly. If they hesitate, that is worth noting, though it does not mean the care is poor.

Where AI radiology is heading

Expect AI to show up in more of your care, often invisibly. Radiology groups are testing AI that drafts reports, compares current and prior scans automatically and flags incidental findings that need follow-up. Some European programs are studying whether AI can safely replace 1 of the 2 human readers in screening. In the US, the FDA continues to add AI-enabled devices to its public list, and the agency has issued guidance on how makers can update AI models after clearance.

The same ambient and generative AI tools spreading through clinics, such as AI medical scribes, are also reaching radiology reading rooms. Data quality and oversight, covered in our guide to healthcare data analytics, will decide how well they work.

Conclusion

AI in radiology is best understood as a fast, tireless assistant. It flags, sorts and measures, and the strongest trial evidence so far, from mammography screening in Sweden, suggests it can help find more cancers while easing radiologists’ workload. A licensed radiologist still reviews your images and signs the report. Your job as a patient is simple: keep your screenings, ask whether and how AI is used, and talk through any result you do not understand with your clinician.

Sources

• FDA: Artificial Intelligence-Enabled Medical Devices list — public list of authorized AI-enabled devices
• Lund University: AI support in breast cancer screening leads to fewer missed cancer cases — MASAI final results published in The Lancet, 2026
• FDA: Final Rule to Amend the Mammography Quality Standards Act — breast density notification, enforced September 10, 2024
• FDA: Understanding Breast Density — why dense tissue matters on mammograms