This Week in Radiology — Sep 26, 2026

This AudioScholar briefing is generated by artificial intelligence for healthcare professionals and trainees. It is not medical advice.

Welcome to This Week in Radiology. This week we're covering 10 notable papers spanning screening in asymptomatic people, cardiovascular imaging from photon-counting CT to opportunistic biomarkers, and questions of safety and new tracers across interventional and molecular imaging. Let's dive in.

We start with screening, where three papers pull in interestingly different directions. In Nature Medicine, Zhou and colleagues report EAGLE, a deep learning model that detects oesophageal cancer and precancerous lesions on noncontrast chest CT, a task the authors frankly describe as historically considered impossible given how readily the oesophagus collapses and how subtle early lesions are [1]. The model was trained on roughly 6,800 patients at two centres and then validated across twelve centres in three countries, involving more than eighty thousand people in both opportunistic and population screening settings. In the external multicentre test cohorts, specificity was about 98 and a half percent with sensitivity around 90 percent for frank cancer, though only about half for precancerous lesions. Performance held up on low-dose CT, which raises the possibility of piggybacking oesophageal detection onto lung cancer screening programmes, and in prospective hospital validation the positive predictive value was just over 40 percent. That is an unusually large and geographically diverse validation for a radiology artificial intelligence tool, but the abstract is clear that the analyses suggesting improved screening efficiency from endoscopic referral are exploratory, and all of the population data come from a small number of health systems, predominantly in China.

Pushing the other way, an American Journal of Roentgenology study from Joo and colleagues asks what happens when you take upfront prostate MRI screening, validated largely in North American and European trial settings, and deploy it in a real-world Korean health check-up programme [2]. Among 618 asymptomatic men with a median prostate specific antigen of about one, roughly one in seven was scored PI-RADS three or above, but clinically significant cancer turned up in only nine men, about one and a half percent. When the authors simulated a conventional pathway in which MRI followed only a prostate specific antigen above three, that strategy would have avoided ninety percent of the MRI examinations and about seven in ten biopsies that found no significant cancer, at the cost of missing a single case, and in the subset of men who would have qualified for the PROSA randomised trial it missed none at all. The authors conclude this supports continued prostate specific antigen gating in this population, and their broader point is the one worth carrying: population-level differences matter before trial results are extrapolated. That caution frames the AJR Expert Panel narrative review on whole-body MRI in asymptomatic individuals, led by Petralia [3]. The panel notes that whole-body MRI is increasingly sold as preventive healthcare to people without cancer risk factors, driven substantially by a commercial market, despite limited evidence of clinical benefit and no guideline endorsement. Against that, they set the harms: workup of indeterminate findings, overdiagnosis of indolent disease, direct and downstream costs, and ethical problems around unregulated advertising and informed consent. Their prescription for the field is prospective trials of clinical benefit and cost-effectiveness, plus identification of intermediate-risk groups in whom targeted screening might actually pay off. One place where structured screening is already well justified is covered in RadioGraphics by Fung and colleagues, reviewing cancer predisposition syndromes in children, which collectively account for perhaps fifteen to twenty percent of paediatric cancers [4]. Surveillance is generally recommended when malignancy risk exceeds five percent within a defined age window, and the review, built on the 2023 American Association for Cancer Research workshop recommendations, emphasises radiation-free modalities, ultrasound and MRI, with whole-body MRI increasingly folded into protocols for syndromes with a broad tumour spectrum.

Turning to cardiovascular imaging, European Radiology brings a practical technology question. Lacaita and colleagues retrospectively compared photon-counting CT coronary angiography with conventional energy-integrating detector CT in 141 consecutive patients with coronary stents, using invasive angiography as reference [5]. The headline is image quality: with photon-counting CT, only about one percent of stents were nondiagnostic, against nearly half with the older detector technology, and around seven in ten stents were graded excellent versus under a quarter. For detecting in-stent restenosis of fifty percent or more, per-patient accuracy was about ninety percent with sensitivity of one hundred percent and, importantly, a negative predictive value of one hundred percent, while positive predictive value was moderate, under sixty percent per stent. So this is a single-centre retrospective series suggesting photon-counting CT can reliably exclude significant in-stent restenosis, while positive calls still tend to need confirmation. Complementing that, a RadioGraphics review from Sorin and colleagues surveys imaging biomarkers for atherosclerotic cardiovascular disease screening, reaffirming coronary artery calcium scoring as the established, guideline-endorsed tool for reclassifying borderline and intermediate risk, while describing plaque burden and composition, pericoronary adipose assessment, and opportunistic findings such as breast arterial calcification, hepatic steatosis and sarcopenia as promising but still requiring standardisation and outcome-based validation [6]. The Journal of the American College of Radiology gives us the real-world counterpart. Groner and colleagues reviewed baseline low-dose CT reports from more than 2,300 lung cancer screening participants in a New York City health system and found emphysema in roughly three quarters and coronary artery calcium in about two thirds [7]. Pulmonary care within twelve months roughly doubled after screening, and cardiovascular care rose from just over forty percent to about sixty percent. Social determinants mattered: lower neighbourhood income and greater food insecurity were each associated with roughly fifty percent higher odds of emphysema, and among participants with coronary calcium, lower area educational attainment was associated with lower odds of receiving cardiovascular follow-up. This is retrospective and, as the authors state plainly, it remains unknown whether the extra utilisation improves outcomes.

Finally, three papers on safety and emerging applications. In the American Journal of Roentgenology, Mabud and colleagues used a national commercial claims database covering nearly sixty thousand women to compare thirty-day safety of uterine fibroid embolization against myomectomy, with the crucial twist of separating myomectomy by approach [8]. Against abdominal myomectomy, embolization roughly halved the odds of a major adverse event, two and a half percent versus about five percent, and cut readmissions substantially. Against laparoscopic myomectomy the advantage was smaller, and against hysteroscopic myomectomy embolization fared worse, with roughly double the odds of a major adverse event and of readmission. Emergency department visits were higher after embolization across all comparisons, and when all myomectomy subtypes were pooled, no difference emerged at all, which is precisely the authors' point: pooling obscures clinically meaningful differences in both directions. In Investigative Radiology, Ludwig and colleagues report preclinical work on gadolinium-based contrast agents, using a mouse model of autoimmune encephalomyelitis plus organotypic hippocampal slice cultures [9]. Neuroinflammation increased hippocampal gadolinium accumulation for both agents tested, but long-term retention out to forty days was seen only with the linear agent gadopentetate. Ex vivo, however, both the linear and the macrocyclic agent at biologically relevant, sub-cytotoxic concentrations reduced mitochondrial length and area and altered motility in neurons, with effects amplified under inflammatory conditions. This is animal and tissue-culture work at cumulative doses far beyond clinical exposure, so it does not speak directly to patient risk, but it does propose a mechanism worth watching in neuroinflammatory disease. And rounding out the week, a Special Series review in the American Journal of Roentgenology from Enke and colleagues makes the case for fibroblast activation protein-directed PET beyond oncology, since the target tracks tissue remodelling, chronic inflammation and fibrosis across cardiac, vascular and rheumatologic disease [10]. The authors are careful to say the supporting evidence is early-stage and largely exploratory.

If you only have time for one paper this week, make it the Nature Medicine EAGLE study [1]. It is the largest external validation yet of an artificial intelligence tool reading a target that radiologists have essentially written off on noncontrast CT, and it reopens the question of what else routine chest imaging is silently carrying.

Here is what this week's evidence adds up to. First, artificial intelligence on existing noncontrast CT can detect oesophageal cancer with high specificity across multiple countries, though detection of precancerous disease remains around half and the efficiency claims are exploratory. Second, screening strategies do not travel unchanged between populations: in Korean men at health check-up, upfront MRI was inefficient and prostate specific antigen gating performed at least as well, while whole-body MRI in unselected asymptomatic adults still lacks evidence of benefit and lacks guideline support. Third, photon-counting CT largely solves the nondiagnostic stent problem in a single-centre series, with excellent negative predictive value but only moderate positive predictive value. Fourth, opportunistic biomarkers on lung cancer screening CT are extremely common and are followed by more cardiopulmonary care, but whether that improves outcomes is unresolved, and follow-up appears unevenly distributed. And fifth, comparative safety for fibroid therapy depends entirely on which myomectomy you are comparing against, with embolization favourable against abdominal surgery and less so against hysteroscopic.

That's your roundup for This Week in Radiology. The full transcript and references are available on the episode page in your AudioScholar library. This is an AI-curated summary — for clinical decisions, always consult primary sources and current guidelines. 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 free 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. Always review the original source materials.