Every imaging center has this patient. She needs the scan, she scheduled the scan, and somewhere between the car and the front desk she read a number about millisieverts that made her hands cold. The number was probably true — when it was collected.
The benchmarks behind most of what patients find — and honestly, behind a lot of what gets said in consent conversations — trace to national data gathered around 2014. Ten years of scanner hardware, reconstruction software and protocol work have happened since. Until this week, nobody had re-measured at national scale.
On Monday, Radiology published the re-measurement: Kalpana Kanal and colleagues analyzed about 5.2 million adult CT acquisitions from 592 U.S. facilities, all performed in 2025, across the ten most common adult CT exam categories, and set them against the 2014 benchmarks.
A diagnostic reference level is a benchmark, not a dose your patient receives — it’s the 75th-percentile marker facilities compare their own protocols against. When the benchmark itself falls by a fifth, it means the whole national distribution moved, not one good machine.
The pattern inside the numbers is as useful as the headline. Torso exams — chest, abdomen-pelvis, and the combined studies — dropped the most. Neck and cervical spine landed in the middle. Head barely moved.
That pattern is not random. The big torso gains line up with where the last decade’s technology actually went to work — iterative and machine-learning reconstruction lets you pull a diagnostic image out of less signal, and torso protocols had the most room to give. Head CT was already running tighter margins, and it shows.
The authors are careful about causation and so am I: the study measured the drop, it did not prove what caused it. Their own framing is that the reductions are consistent with the decade’s changes in reconstruction and protocols — measured result, plausible mechanism, stated in that order.
The magnitude and consistency of these reductions across examination categories suggest that such advances have translated into measurable, population-level dose reductions in routine clinical imaging in the United States.
Kept honest. The sample is real-world but not a census: the data comes from facilities on the Imalogix dose-management platform — 592 of them — and facilities that run dose-optimization software may sit ahead of the national average, which would make these numbers flatter the field slightly. The study measured the decline, not the cause — the technology explanation is the authors’ consistent-with reading, not a proven mechanism. DRLs are 75th-percentile benchmarks, not individual patient doses; no patient should be told “your dose fell 22%.” And a disclosure about my own sourcing: the journal’s site blocks automated access, so I could not read the full paper today. Every figure here was cross-checked against two independent trade reports published this week — Radiology Business (July 30) and AuntMinnie (July 29) — which agree on every number used above, and the paper’s DOI is real and current: 10.1148/radiol.260322, with an accompanying Radiology editorial titled “Good News about CT Doses.” Where the two reports diverged on anything, that figure stayed out of this piece.
Source: Kanal et al., Radiology (RSNA), published July 28, 2026 — doi 10.1148/radiol.260322 — U.S. diagnostic reference levels for the ten most common adult CT examinations, 2025 data: ~5.2 million acquisitions, 592 facilities. Disclosure per house rules: pubs.rsna.org blocks automated fetching, so the paper could not be read directly today; per the WAF-block rule, all figures were cross-checked across two independent reports — Radiology Business, July 30 and AuntMinnie, July 29 — which agree exactly on every figure used: 5.2M exams, 592 facilities, DRLs down 22% on average, chest with contrast −31%, chest without −27%, head without contrast −4%, torso categories largest / neck intermediate / head minimal. The accompanying editorial is “Good News about CT Doses,” doi 10.1148/radiol.261781. Any figure the two reports did not both carry was excluded.
The patient who's afraid of her CT scan googled the radiation numbers from your parking lot. What she found was benchmarked in 2014. On Monday, the field re-measured itself. Kanal and colleagues, in Radiology: about 5.2 million adult CT acquisitions, 592 U.S. facilities, all performed in 2025, across the ten most common adult CT exam categories — set against the national benchmarks collected a decade ago. Diagnostic reference levels fell an average of 22%. Chest with contrast: down 31%. Chest without: down 27%. Head without contrast: down 4%. Yes, I'm leading with the weak number too. More on that in a second. What a DRL is, for anyone outside the physics room: it's the 75th-percentile benchmark facilities compare their protocols against — not a dose any individual patient receives. Which is exactly why this matters: when the BENCHMARK falls by a fifth, the whole national distribution moved. Not one good machine. The field. The pattern inside the numbers: torso exams dropped most, neck and c-spine landed in the middle, head barely moved. That tracks with where a decade of iterative and machine-learning reconstruction actually went to work — torso protocols had room to give, head was already running tight. The authors are careful about causation and so am I: the study measured the drop. It didn't prove the cause. Their framing is that the reductions are "consistent with" the decade's technology changes. Measured result, plausible mechanism, in that order. Three things I'd do with this if I ran an imaging center: 1. Read the consent-room script this week. If the dose comparisons your techs use cite decade-old figures, they now overstate chest CT dose by a quarter to a third against the national benchmark. Don't promise a number. Just stop quoting a stale one. 2. Have your physicist benchmark protocols against the 2025 DRLs. A protocol comfortably under the old reference levels may sit above the new ones. That's literally what DRLs are for. 3. Write the patient-facing version with dignity. The claim is "the field re-measured at national scale and doses fell substantially." It is not "CT is safe now," and it's not a promise about anyone's individual exam. The honest version builds more trust — especially if you volunteer the head CT number before anyone asks. Kept honest: the data comes from 592 facilities on a dose-management platform, and facilities running optimization software may sit ahead of the national average. And the journal's site blocks automated access, so I verified every figure across two independent trade reports that agree exactly — the DOI is 10.1148/radiol.260322, published July 28. The fear is running on 2014 data. The machines moved on. Now the conversation can too.
A finished, narrated animation of this edition came with your email (MP4, 1080×1350 — sized for LinkedIn and Instagram). Post it as-is.
FIRST SOURCE · one verified original-source finding, composed for one reader · this edition: radiology — written for the imaging practices whose hardest conversations happen before the scan, not after it