AI in Radiology Imaging: Why Patients Now Expect Full Transparency and Consent

The Number That Should Stop You Mid-Scroll

Nearly 100% of respondents want disclosure. Let that land for a second.

This isn’t a niche concern from a tech-anxious fringe. It’s a broad, cross-demographic expectation from the people actually lying in the scanner. And yet, disclosure of AI tools in radiology reporting is currently not mandatory in most jurisdictions.

That gap — between what patients expect and what systems require — is where the real story lives.

Consent: Written, Verbal, or Just… Assumed?

Of the patients who wanted to be informed, 53% preferred written consent versus 34% who were fine with verbal acknowledgment. The remaining slice presumably hadn’t decided, or didn’t mind either way.

The researchers, led by Dr. Hayley Briody at Beaumont Hospital in Dublin, flag the obvious tension here. Written consent is thorough. It’s also slow. In a busy imaging department processing hundreds of scans daily, adding a consent workflow for every AI-assisted read isn’t trivial.

The U.S. Blueprint for an AI Bill of Rights gestures toward a “right to notice and explanation” — but stops short of defining whether that means a standardized notification or full informed consent. Helpful. Very helpful.

Who’s Responsible When AI Gets It Wrong?

Here’s where it gets philosophically interesting.

When asked who bears fault if an AI-assisted read produces an incorrect result, 64% of respondents pointed at both the radiologist and the technology. Not just the doctor. Not just the algorithm. Both.

That’s a sophisticated answer, actually. Patients aren’t naively offloading blame onto the machine, nor are they letting the human off the hook. They’re holding the system accountable — the whole system.

Dr. Briody and co-authors interpret this as a preference for transparency, and it’s hard to disagree. Patients seem to understand, intuitively, that AI in clinical settings is a collaboration. They just want that collaboration acknowledged.

Why This Research Matters Beyond Radiology

Most prior literature on clinical AI perception focused on providers and medical students. Patient voices were largely absent from the conversation — which is a peculiar omission given that patients are the ones most directly affected.

This study starts to close that gap. It’s limited, as the authors readily admit: single-language survey, unknown response rate, potential selection bias from waiting-room recruitment. But it’s a starting point, and a directionally clear one.

The authors also flag an underexplored question worth watching: do patient attitudes shift when AI is used for secondary purposes, like triaging radiology requests rather than direct image interpretation? That’s a meaningfully different use case, and the answer probably isn’t the same.

What This Means for AI Tool Builders and Healthcare Operators

If you’re building or deploying AI in clinical environments, this research is a quiet but firm nudge.

Transparency isn’t just an ethical nicety — it’s becoming a patient expectation baked into trust. Tools that operate invisibly, even when they perform well, are accumulating a credibility deficit. And when something goes wrong, that deficit compounds fast.

The governance question is equally pointed. Involving patient representatives in AI oversight isn’t a PR move. It’s what the people using these systems are actually asking for.

The Takeaway

Patients aren’t afraid of AI in radiology. They just want to know it’s there.

That’s a reasonable ask. And the fact that current frameworks haven’t fully answered it — despite near-universal patient preference — suggests the industry is still catching up to the people it serves. The tools are advancing fast. The consent infrastructure, less so.

Observe the gap. Then close it.