This week in Radiology
Nov 21, 2026
Medical Imaging and Pediatric and Adolescent Brain Tumor Risk.
In about 3.7 million children, brain tumor risk rose with cumulative imaging radiation dose, with roughly one excess tumor per 1,266 head CT examinations.
JAMA · 2026 · PubMed
This week’s highlights
Read the weekly briefing
This AudioScholar briefing is generated by artificial intelligence from PubMed abstracts for healthcare professionals and is not medical advice.
This week: childhood imaging radiation and brain tumor risk rising with dose, thrombectomy outcomes in very late-presenting stroke, and early postmarket safety reports for histotripsy liver ablation.
Welcome to This Week in Radiology. We're covering 10 papers this week: 3 in depth, 4 in brief, 3 quick signals.
In JAMA, Miglioretti and colleagues report a retrospective cohort of children born between 1996 and 2016 and enrolled from birth in six United States health care systems or in Ontario, Canada [1]. The cohort included about 3.7 million children, followed for a mean of about twelve years, and 1,536 brain tumors were diagnosed. Brain dose from imaging was lagged by two years. A single head CT delivered a mean of about 66 milligray to the brain. Brain tumor risk rose with cumulative dose, reaching close to four times the risk of unexposed children at 150 milligray or more, and the association was stronger for aggressive histologies. Exposure comparable to one head CT was linked to roughly one excess brain tumor per 1,266 head CT examinations, and the authors estimate that about a tenth of brain tumors in the cohort were attributable to imaging radiation. The main limitation is the observational design, so residual confounding by the reasons children were scanned cannot be excluded. The study adds a large, direct dose-response estimate from modern imaging, and the authors conclude it supports careful clinical justification and dose minimization when pediatric imaging with ionizing radiation is indicated.
In Radiology, Yi and colleagues studied thrombectomy far beyond the usual time window [2]. This retrospective cohort drew on prospectively maintained registries from eight stroke centers in China and included 308 patients with intracranial carotid, M1 or M2 occlusion, a stroke scale score of 6 or more and an ASPECTS of 3 or more, presenting between 24 and 96 hours after onset. Of these patients, 205 underwent endovascular treatment and 103 received best medical management. Functional independence at ninety days was reached by about half of the treated patients, compared with just over a quarter of the medically managed patients. After weighting for baseline differences, the odds of independence were roughly four times higher with thrombectomy. Symptomatic intracranial hemorrhage and ninety-day mortality did not differ significantly between groups. The key limitation is that treatment was not randomized, so the patients chosen for thrombectomy may have differed in ways the adjustment did not capture, and the sample is modest and from a single country. The study adds observational support for treating selected late presenters, a question that only randomized trials can settle.
Also in Radiology, Kundu and colleagues analyzed the FDA MAUDE adverse event database for histotripsy during its first two postmarket years, comparing it with microwave and radiofrequency ablation over equivalent early windows [3]. There were 71 histotripsy reports and 435 reports for the thermal technologies. Serious injury appeared in 86 percent of histotripsy reports against 20 percent for thermal ablation, and death in 14 percent against 2 percent. The reporting odds of serious injury were roughly fourteen times higher for histotripsy, with excess reports of renal and urinary complications, hemorrhage and vascular injury, and gastrointestinal or abdominal complications. Exploratory estimates using modeled procedure volumes also pointed to higher relative risk. The central limitation is that MAUDE is a passive reporting system without a true denominator, and the comparison spans different eras and stages of technology maturity, which the authors acknowledge may have shaped reporting. The analysis adds an early safety signal for a newly cleared liver ablation technology, and the authors call for prospective evaluation.
Now, four in brief.
Staying with stroke, the American Society of Functional Neuroradiology published a state-of-practice document in the American Journal of Neuroradiology, led by Sriwastwa, on minor stroke with anterior large vessel occlusion [4]. Despite a stroke scale score of 5 or less, nearly a third of these patients have poor functional outcomes, often from early neurological deterioration. The document names CT angiography as the primary tool for occlusion detection and collateral assessment, with CT and MR perfusion helping predict deterioration. It also notes that AI detection tools may miss proximal M2 and non-terminus carotid occlusions. Like the late-window registry, it addresses patients outside the landmark trials, and the benefit of thrombectomy remains uncertain until the STEP and MOSTE trials report.
In Radiology, Wang and colleagues ran a secondary analysis of a prospective multicenter observational study of 1,583 patients with combined pulmonary fibrosis and emphysema [5]. A commercial AI workstation quantified fibrosis and emphysema on high-resolution CT, and patients were classed as fibrosis-dominant when the fibrosis percentage equaled or exceeded the emphysema percentage. Just over half of the patients were fibrosis-dominant. Over a median of about two and a half years, 513 deaths occurred, and the fibrosis-dominant phenotype carried roughly a third higher mortality after adjustment. The cohort is large and prospective, though the classification relies on one software platform and a simple threshold.
In the Journal of the American College of Radiology, Srour and colleagues systematically reviewed standalone commercial AI for pulmonary nodule detection on chest radiographs [6]. Six studies covered twelve evaluations of ten products. Performance varied: some products beat radiologists, others did not differ, and one produced about twice the false-positive rate of individual readers. Even on a single shared dataset, area under the curve ranged from about 0.79 to 0.93, against about 0.81 for radiologists. All studies were retrospective or reader studies with common bias concerns, and the authors judge the evidence insufficient for class-wide standalone deployment.
In European Radiology, Vanzulli and colleagues compared conventional and delayed hepatobiliary-phase gadoxetate MRI for bile leaks in a single-center retrospective study of 35 patients, mostly after liver transplant or cholecystectomy [7]. Delayed images, acquired at a median of about an hour, raised sensitivity from about a quarter to just over 80 percent, with specificity at 100 percent at both timepoints. Reader agreement was excellent. The result is consistent and clinically plausible, but the sample is small, from one referral center, and relies on a composite reference standard.
Three quick signals.
Also on the liver, the ACR Appropriateness Criteria for chronic liver disease have a 2026 update in the Journal of the American College of Radiology, led by Fung [8]. It rates imaging for hemochromatosis, steatosis and fibrosis using RAND/UCLA and GRADE methods, with expert opinion filling gaps where evidence is lacking or equivocal.
In RadioGraphics, Nadeem and colleagues offer an intent-based framework for personalized yttrium 90 radioembolization dosimetry [9]. They cite prospective evidence that personalized dosimetry improves tumor response and survival, and they present a decision algorithm linking curative or palliative intent to model choice and dose limits. This is a narrative review rather than new data.
In Investigative Radiology, Zeilinger and colleagues compared photon-counting CT with flat-detector CT after cochlear implantation in 28 scans [10]. Photon-counting image quality was noninferior, with fewer metal and motion artifacts, at an effective dose roughly nine times lower. It is a small retrospective single-institution series with years between paired scans.
Stepping back across this week's papers, much of the evidence comes from retrospective cohorts, registries and passive reporting systems, with the largest dataset of the week still observational. Several of the clinical questions raised, from late and minor stroke treatment to new ablation technology and standalone AI, sit in territory where randomized or prospective data are absent or pending, and consensus documents and reviews are filling that space in the meantime.
If you only have time for one paper this week, make it the pediatric imaging and brain tumor cohort [1]. It gives a large, dose-graded estimate of cancer risk from the scans radiologists protocol every day, and it reopens the question of how pediatric CT use and dose are justified.
That's your roundup for This Week in Radiology. The full transcript and references are available on the episode page. See you next week.
If this weekly briefing is useful, follow the show in your podcast app so new episodes arrive automatically. And think of one colleague, in any specialty, who never has time to keep up with the literature. Tell them about AudioScholar: a ten-minute weekly for every specialty, to listen to in any podcast app, or to read at audioscholar dot C C.
This is an automated summary generated by artificial intelligence, which can make mistakes. Use it to choose which papers to read in full.
Get Radiology every week
By email, Spotify, Apple or any podcast app.
Spot something worth flagging?