CT radiation dose is down, and now there are numbers to prove it
Radiation dose from CT scanning has dropped substantially in the United States over the past decade. A nationwide survey published in Radiology analyzed 5,234,285 CT exams performed on 2,802,416 adults in 2025 and compared them with the 2014 benchmark data: diagnostic reference levels fell 21.8% and achievable doses fell 9.3% across ten common adult exam categories. After a stretch of alarming headlines about medical radiation, the curve is finally pointing the right way.
How the survey was built
The work was led by Kalpana M. Kanal, PhD, of the Department of Radiology at the University of Washington School of Medicine in Seattle, with David Frush, Eric Gingold and colleagues. It is a retrospective, observational analysis pooling dose data from 592 U.S. facilities, processed with a modernized, acquisition-level and size-normalized framework.
The comparison works because the baseline is well defined. In 2014 the same group published the survey that became the field’s reference point, covering 1.3 million exams at 583 CT sites. The current sample is roughly four times larger, which makes the percentiles far more stable. “Using a modernized, acquisition-level, size-normalized framework, substantial reductions in U.S. adult CT radiation dose were demonstrated over the past decade,” Kanal and colleagues wrote.
DRL versus AD: why two different numbers circulate
One point deserves clearing up, because coverage of the paper has quoted different figures. The diagnostic reference level (DRL) sits at the 75th percentile of the dose distribution — an informal ceiling above which a department should audit its protocol. The achievable dose (AD) sits at the 50th percentile, the median, and works as an optimization target. The two moved at different speeds: DRLs dropped 21.8%, while ADs dropped 9.3%. In plain terms, the departments that were far off the curve improved a great deal, and the median improved too, just less dramatically. Headlines saying “22% lower” are quoting the DRL; those saying 9.3% are quoting the median.
The metrics themselves are worth a line. Volume CT dose index, $\mathrm{CTDI}_{vol}$, is reported in mGy and describes average dose in a reference phantom for a given acquisition. Dose-length product scales with the irradiated extent:
$$\mathrm{DLP} = \mathrm{CTDI}_{vol} \times L$$
where $L$ is scan length in cm — which explains why DLP fell 20% while $\mathrm{CTDI}_{vol}$ fell further. Trimming scan coverage is part of optimization, but not the largest part of it.
Where dose fell most, and where it barely moved
The reductions were not uniform, and that is what makes the paper useful for anyone tuning protocols. The largest $\mathrm{CTDI}_{vol}$ DRL reduction came in contrast-enhanced chest CT, down 31.2%. Non-contrast chest CT followed at 26.7%, non-contrast cervical spine at 25% and contrast-enhanced neck CT at 15%.
The outlier is non-contrast brain CT: just 3.5% lower in ten years. That is not a footnote. Head CT is among the highest-volume exams anywhere, particularly in the emergency department, and the flat line suggests neuro protocols have received far less optimization attention than chest and abdomen. If there is an obvious frontier for the next decade, this is it.
In an accompanying editorial, Cynthia H. McCollough, PhD, professor of medical physics and biomedical engineering at the Mayo Clinic in Rochester, translated the new percentiles into practice: against the benchmarks the American College of Radiology recommended back in 2004, the new DRLs could mean roughly 20 mGy less for head CT and 14 mGy less for routine abdominal CT. “This is a remarkable achievement for the CT community, particularly for routine scanning of the abdomen,” she wrote.
What is driving the improvement
The retrospective design cannot establish causality, and the authors say so. They do point to three likely drivers: hardware advances, better protocol design and artificial intelligence. On the scanner floor that means automatic tube current modulation, automatic kV selection, iterative reconstruction and, more recently, deep learning reconstruction, which keeps noise acceptable in acquisitions that would once have been considered underdosed. “The magnitude and consistency of these reductions across examination categories suggest that such advances have translated into measurable, population-level dose reductions,” the authors note.
It follows a pattern we keep seeing in imaging: the constraint is rarely the technology on the market but the standardization of the people using it — the same dynamic behind the evidence that clinical knowledge cuts low-value imaging.
What it means for an imaging department
For the medical physicist and the lead technologist, the message is operational: your dose targets have aged. A department benchmarking itself against 2014 references — or worse, 2004 ones — may be congratulating itself on performance that now sits in the upper third of the U.S. distribution. Revisiting the alert thresholds in your dose-monitoring platform, exam type by exam type, is the most immediate action item here.
The second message concerns internal variability. Much of the DRL drop comes from departments that retired outlier protocols, not from a universal technology leap. That is reproducible anywhere: comparing $\mathrm{CTDI}_{vol}$ and DLP across your own scanners, for the same exam type, usually surfaces gaps that are hard to defend clinically.
And justification precedes optimization. No amount of dose tuning redeems an unnecessary exam — a debate that also runs through the handling of incidental findings in CT lung screening. The lowest dose is still the scan that did not need to happen.
Context and next steps
The numbers land as a counterweight to a noisy cycle, including the contested 2025 paper estimating that routine CT could cause more than 100,000 cancers a year in the U.S. This survey does not dismiss concern about medical radiation; it shows the trend line moving in the right direction and hands radiologists real figures for the patient who asks about risk.
The limits are clear. These are U.S. data from facilities that voluntarily join dose registries, which selects for better-organized institutions, and they do not transfer automatically to health systems elsewhere. The method does transfer, though, and national reference-level discussions get easier when there is an updated international target to argue with. Screening programs are further along that road, as the real-world data on CT lung screening uptake and mortality showed. For everyday diagnostic CT, the work is just starting.
Source: The Imaging Wire e Diagnostic Imaging — original study: Kanal KM, Frush DP, Gingold E, et al. Radiology 2026;320(1). DOI: 10.1148/radiol.260322.
