This Week in Radiology — Sep 5, 2026

Welcome to This Week in Radiology. This week we're covering 10 notable papers spanning cancer screening and surveillance, neurovascular imaging from stroke triage to endovascular complications, and the practical infrastructure of artificial intelligence in the reading room. Let's dive in.

We'll start with screening and surveillance, where two RadioGraphics reviews push in the same direction: stop treating screening as a single test for a single disease. Fowler and colleagues synthesise the evidence on hepatocellular carcinoma surveillance in patients at high risk, and the message is that semiannual ultrasound with alpha-fetoprotein remains the standard, but it is not equally good for everyone [1]. Sensitivity falls off in patients with obesity, hepatic steatosis, and advanced cirrhosis, and the Liver Imaging Reporting and Data System 2024 guidance now folds examination quality directly into the management algorithm — so if your ultrasound is visually limited, that finding itself should drive the next step rather than being buried in a caveat sentence. For those patients, biomarker panels and abbreviated MRI protocols show higher sensitivity and favourable cost-effectiveness when applied to appropriately risk-stratified groups. The authors argue explicitly for moving away from one-size-fits-all surveillance, while acknowledging that uptake in the real world remains poor for patient, provider, and system reasons. Alongside that, Singh and colleagues make the case for opportunistic screening on low-dose CT for lung cancer [2]. In the National Lung Screening Trial, roughly a third of screening examinations showed abnormalities other than lung cancer, and the paper offers a practical framework for emphysema, interstitial lung abnormalities, coronary artery calcification, thoracic aortic aneurysm, pulmonary artery enlargement, low bone density, and upper abdominal findings — with appropriate use of the Lung CT Screening Reporting and Data System "S" modifier and structured reporting. These are conditions that carry substantial morbidity if unrecognised, and the scan has already been acquired at no extra dose or patient burden.

That leads naturally to the New England Journal of Medicine review by Callister and Silvestri on pulmonary nodules, which is the downstream problem screening creates [3]. The framework is familiar but worth reinforcing: solid nodules stable for two years are considered benign, whereas subsolid nodules need a longer period of stability before you can say the same. Subsolid nodules grow more slowly but carry a higher malignancy risk, particularly when a solid component appears or progressively enlarges. Comparison with prior imaging remains the single most valuable step, risk prediction models guide solid nodule management, and the pathway runs from CT surveillance for low-risk lesions, to PET-CT or biopsy for intermediate risk, to resection for selected high-risk nodules. The recurring theme is balance — timely diagnosis in the patient who has cancer against avoiding invasive procedures in the many who do not.

Moving to neurovascular imaging, where three papers address how we grade, how we follow up, and what goes wrong. In the American Journal of Neuroradiology, Elek and colleagues pooled 48 studies and nearly 4,900 patients to ask whether the posterior circulation ASPECTS score predicts outcome differently depending on the modality it is measured on [6]. Patients with favourable 90-day outcomes had consistently higher scores, and each one-point drop in the score was associated with roughly a fifty percent increase in the odds of an unfavourable outcome. Diffusion-weighted imaging, non-contrast CT, and CT angiography source imaging all performed in the same direction with no statistically significant difference between them — but the authors are careful, and so should we be: the absence of a difference is not proof of equivalence, and they explicitly warn against using the score as a stand-alone threshold for deciding who gets thrombectomy. Complementing that, Güntert and colleagues review leptomeningeal collateral grading in the same journal, and their conclusion is one of unresolved variability [9]. Digital subtraction angiography remains the reference standard but is invasive and largely restricted to patients already going to thrombectomy, while the many proposed single-phase CT angiography grading systems differ considerably in interrater reliability and standardisation, with no consensus on which best captures collateral anatomy. Automated, reproducible collateral scoring is the plausible fix, but it has not yet entered routine practice.

On the interventional side, Sasaki and colleagues analysed more than 53,000 procedures in the Japanese nationwide neuroendovascular registry to benchmark iatrogenic vessel dissection [7]. Overall incidence was low, about six or seven per thousand procedures, but the spread across procedure types is what matters for consent and planning — from roughly one in a thousand in vasospasm treatment up to about three and a half percent in intracranial angioplasty and stenting. Dissection clustered around flow diverters, angioplasty and stenting, and lesions requiring balloon dilation or stent deployment, and was more frequent in emergency cases and under local anaesthesia. Reassuringly, among nearly 300 isolated dissections, close to ninety percent OF CASES showed no functional decline at 30 days. The harm came not from the dissection itself but from secondary intracranial haemorrhage or infarction, and at intracranial sites deaths occurred only after haemorrhage. An independent association with severe disability or death was seen only in aneurysm treatment. Staying with flow diverters, Gao and colleagues in the European Journal of Radiology assessed photon-counting detector CT for non-invasive follow-up in 30 patients with 32 aneurysms [8]. Stent lumen visibility scored substantially higher on photon-counting CT than on either baseline digital subtraction angiography or flat-panel VasoCT, with excellent interobserver agreement, and diagnostic confidence approached that of catheter angiography while remaining below it. Radiation dose was under one millisievert, higher than localised angiography because of whole-brain coverage. This is a small retrospective cohort, and the authors are clear that prospective validation of diagnostic accuracy for in-stent stenosis is still needed before it replaces catheter follow-up.

Two more papers, one cardiac and two on artificial intelligence infrastructure. Menard and colleagues, also in the European Journal of Radiology, looked at 159 patients on cardiac MRI — 55 with cardiac amyloidosis and matched groups with hypertrophic and dilated cardiomyopathy — and found that late gadolinium enhancement of the interatrial septum is far more common in amyloidosis [5]. Anterior septal enhancement was present in over eighty percent OF AMYLOID CASES versus about fifteen percent in hypertrophic cardiomyopathy. Critically, intense enhancement — defined as equal to or greater than ventricular enhancement — was 98 percent specific for amyloidosis. Septal thickness itself did not discriminate. This is retrospective and single-cohort, but it is a free observation to add to your routine cardiac MRI checklist. On the AI side, Schmidt and colleagues in the Journal of the American College of Radiology describe the technical workflow behind Assess-AI, the first national imaging AI registry, using large language models to extract findings from radiology reports for nine use cases including intracranial haemorrhage and pulmonary embolism [4]. Agreement with report-derived reference labels was very high, but the authors state plainly that those cohorts also informed prompt refinement, so independent accuracy and clinical utility remain unestablished. Complementing that plumbing work, Wen and colleagues in European Radiology present RAPID, a deep learning framework that labels CT series anatomically from the topogram rather than relying on inconsistent DICOM text metadata, with strong classification and landmark detection performance internally and on external validation, and radiologist review broadly agreeing with the technical numbers [10].

If you only have time for one paper this week, make it the RadioGraphics review on hepatocellular carcinoma surveillance [1]. It changes how you should report a technically limited surveillance ultrasound and gives you the evidence to recommend abbreviated MRI in the patients where ultrasound predictably underperforms.

Here are the key takeaways from this week in Radiology. First, surveillance ultrasound quality is now part of the management algorithm, not a footnote — document it and escalate to abbreviated MRI or biomarkers in patients with obesity, steatosis, or advanced cirrhosis. Second, treat the lung cancer screening CT as a whole-chest risk assessment, and report emphysema, interstitial abnormalities, and coronary calcium using the structured "S" modifier. Third, posterior circulation ASPECTS predicts 90-day outcome consistently across CT, CT angiography, and diffusion imaging, but it should inform, not dictate, thrombectomy selection. Fourth, iatrogenic dissection during neuroendovascular therapy is usually benign — it is the secondary haemorrhage or infarction that drives poor outcomes, and the risk is highest with angioplasty, stenting, and flow diverters. And fifth, add the interatrial septum to your cardiac MRI search pattern: intense late enhancement there is highly specific for amyloidosis.

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 if it saves you time, a quick rating in your podcast app helps other physicians find the show.

This is an automated summary generated by artificial intelligence, which can make mistakes. Always review the original source materials.