This Week in Radiology — Aug 1, 2026

Welcome to This Week in Radiology. This week we're covering 10 notable papers spanning neurointervention and stroke imaging, oncologic imaging and image-guided biopsy, and the practical business of radiology — automation, quality frameworks, and how we communicate results to patients. Let's dive in.

We'll start with neurointervention, where three papers from the American Journal of Neuroradiology map out both consensus and remaining uncertainty. The first is the ARISE II consensus statement on venous sinus stenting [1]. Venous sinus stenosis has emerged as a key contributor to idiopathic intracranial hypertension and is now recognised as the most common cause of pulsatile tinnitus, and as noninvasive venous imaging has improved, we are detecting it far more often. A multidisciplinary roundtable of academic, government, and industry cerebrovascular experts convened to standardise what has until now been a highly variable practice, covering patient selection, diagnostic evaluation including invasive manometry and pressure gradients, procedural technique, and post-treatment follow-up. The headline is that expert consensus supports venous sinus stenting as a safe and effective therapy in appropriately selected patients. For the practising neuroradiologist, this is the document to anchor your local protocol to, and it also flags where the evidence is thinnest and further investigation is needed. Also in the American Journal of Neuroradiology, a multi-institutional propensity-matched analysis compared intravenous cangrelor against glycoprotein two-b three-a inhibitors during thrombectomy for middle cerebral artery stroke [3]. Using a research network spanning more than a hundred healthcare organisations, the investigators matched three hundred and two patients in each arm. In the overall cohort, there was no significant difference between the two agents in major bleeding, intracranial haemorrhage, or mortality — the drugs performed comparably. The interesting signal sits in a prespecified subgroup of patients who underwent adjunctive angioplasty or stenting, roughly a hundred per arm, where cangrelor was associated with about half the rate of major bleeding, intracranial haemorrhage, decompressive hemicraniectomy, and functional dependency. But that came with a trade-off: recurrent stroke over the following one to twenty-four months was substantially higher with cangrelor, roughly a third of patients versus about one in nine. This is exploratory, retrospective, and subgroup-driven, so it should not change your default agent, but it does frame a genuine risk-benefit tension — less bleeding up front, more recurrent ischaemia downstream — that argues for individualised antiplatelet selection when you rescue-stent. Rounding out the neurovascular theme, a review in the same journal tackles distal and medium vessel occlusions [7]. The practical message for anyone reading acute stroke studies is that CT angiography frequently misses these occlusions given their anatomical variability, and perfusion-derived time-to-maximum maps have shown significantly higher sensitivity for detection. Artificial intelligence tools are promising here but remain inadequately validated, and thrombectomy for distal occlusions is still debated, with evidence largely confined to single-centre series. If your protocol relies on CT angiography alone for occlusion detection, this review is a prompt to make sure perfusion maps are being reviewed for distal targets rather than treated as an afterthought.

Moving to oncologic imaging, three papers address how we characterise and sample tumours. In European Radiology, a multicentre retrospective study of nearly sixteen hundred peripheral lung lesions asked what a non-diagnostic percutaneous transthoracic needle biopsy actually means, and whether contrast-enhanced ultrasound guidance can reduce those results [2]. Strikingly, about forty percent of biopsies yielded non-diagnostic findings — and those results are not homogeneous. Malignancy risk rose with lesion size, granulomatous inflammation or organising pneumonia was protective, and lesions returning atypical cells suspicious for malignancy carried substantially higher malignancy risk than atypical cells of undetermined significance. Nonspecific benign results carried the lowest risk. Importantly, insufficient sample and outright sampling failure carried no significant difference in malignancy rates, so both deserve the same follow-up seriousness. On technique, contrast-enhanced ultrasound guidance delivered significantly higher sampling success, diagnostic accuracy, and sensitivity compared with conventional ultrasound guidance, and the benefit was concentrated in lesions between three and five centimetres, where necrosis is common and contrast helps you target viable tumour. That is an actionable, size-based rule for selecting your guidance modality. In the Journal of Magnetic Resonance Imaging, a prospective study of one hundred fifty-four women with one hundred fifty-nine breast lesions evaluated MR elastography, addressing the unsettled question of which driver frequency to use [8]. Comparing ten frequency sets, the forty to seventy hertz multifrequency band gave significantly better image quality than all others. Stiffness and viscosity differed significantly between oestrogen-receptor-positive and negative tumours, across Ki-67 levels, T stage, molecular subtypes, and enhancement patterns. Stiffness emerged as an independent predictor of triple-negative disease, with a model combining stiffness and age achieving an area under the curve of about zero point seven four in training and zero point seven in validation — modest discrimination, so this is a stage-two technical validation rather than a clinically deployable biomarker, but it does settle the protocol question for anyone building a breast elastography sequence. Complementing that, the European Journal of Radiology published a pictorial review on VI-RADS and bladder MRI [6]. The authors argue the field's central challenge has shifted from proving diagnostic accuracy for distinguishing non-muscle-invasive from muscle-invasive disease, to implementation — ensuring consistent image quality across scanners, reducing inter-reader variability, and translating VI-RADS categories into actual clinical pathways. They highlight the VI-RADS Quality Score as a way to separate a technically inadequate examination from a genuinely equivocal lesion, which is a distinction that matters enormously for the urologist reading your report. They also cover post-therapy scoring systems for response after neoadjuvant therapy and bladder-preserving treatment, and position near-term artificial intelligence as most useful for scan acceleration, automated quality auditing, and semi-automated segmentation rather than autonomous diagnosis.

The third theme this week is where radiology meets the rest of the health system — automation, systemic disease recognition, and patient communication. In the American Journal of Neuroradiology, a retrospective study of seventy clinical fetal MRI examinations evaluated whether automated brain segmentation from slice-to-volume reconstruction could replace manual two-dimensional measurements [10]. Total reconstruction and volumetric analysis took under nine minutes on average, and segmentation quality was rated good in seventy percent of cases, with about seventeen percent rated poor. Critically, the failures clustered in younger fetuses — mean gestational age around twenty-five weeks versus twenty-nine weeks for the rest — so this pipeline is least reliable exactly where the brain is smallest. Where it worked, automated volumes correlated strongly with the manual linear measurements radiologists already report: supratentorial volume against biparietal and fronto-occipital diameters, cerebellar volume against transverse cerebellar diameter, and brainstem volume against anterior-posterior pons measurement. This is a plausible route to replacing manual calipers, but with a clear caveat about early gestation. From the European Journal of Radiology comes a single-centre retrospective cohort of ninety-five patients with fever of unknown origin who underwent FDG PET/CT and showed a shared bone marrow, spleen, and lymph node hypermetabolic pattern [9]. Forty had adult-onset Still disease and fifty-five did not meet Yamaguchi criteria — and the PET/CT appearances were broadly similar between them, confirming this phenotype is nonspecific. What the imaging did add was stratification: unsupervised clustering separated high from low inflammatory burden, capturing eighty percent of the Still disease patients and forty-two percent of the undiagnosed group, and the high-burden cluster went on to receive significantly different glucocorticoid dose trajectories and more frequent disease-modifying therapy, robust to adjustment for baseline steroid dose, C-reactive protein and ferritin. The authors are explicit that this is exploratory and hypothesis-generating, requiring prospective validation. Finally, a study in the Journal of the American College of Radiology addresses something the Cures Act made unavoidable: when should a patient see their breast biopsy pathology [4]? Among eight hundred thirty-eight women biopsied at a safety-net institution, nearly seven in ten preferred to wait and hear results from a provider rather than receive them immediately by electronic release. Hispanic or Latino patients had roughly double the odds of preferring deferred release, as did patients whose preferred language fell outside English or Spanish. Conversely, patients undergoing MRI-guided biopsy were markedly less likely to defer than those undergoing ultrasound-guided biopsy — plausibly a more informed, previously-navigated cohort. The practical implication is that a one-size-fits-all immediate-release policy does not match what most patients in this setting want, and asking the preference at the time of biopsy is a low-cost intervention. I'd also flag a Radiographics review on thoracic manifestations of obstetric and gynaecologic disease [5], covering peripartum pulmonary oedema, embolic disease, cardiomyopathy, metastatic gestational trophoblastic disease, and the benign entities — Meigs syndrome, thoracic endometriosis, and intravascular leiomyomatosis — that are easy to miss on a chest study when the gynaecologic history isn't provided.

If you only have time for one paper this week, make it the European Radiology study on non-diagnostic lung biopsies and contrast-enhanced ultrasound guidance [2]. It gives you a directly usable risk-stratification framework for the forty percent of biopsies that come back inconclusive, plus a concrete, size-based rule for choosing your guidance modality.

Here are the key takeaways from this week in Radiology. First, a non-diagnostic lung biopsy is not one entity — atypical cells suspicious for malignancy carry high risk, nonspecific benign findings low risk, and insufficient samples deserve the same follow-up as outright sampling failures; consider contrast-enhanced ultrasound guidance for lesions between three and five centimetres. Second, venous sinus stenting now has a consensus framework for selection, technique, and follow-up in idiopathic intracranial hypertension and pulsatile tinnitus. Third, cangrelor and glycoprotein two-b three-a inhibitors performed comparably overall during middle cerebral artery thrombectomy; only in the exploratory rescue-stenting subgroup did cangrelor show less bleeding but more recurrent stroke. Fourth, CT angiography alone under-detects distal and medium vessel occlusions — use perfusion time-to-maximum maps. And fifth, automation is maturing but not autonomous: fetal brain segmentation is good in seventy percent of cases and fails preferentially in early gestation, while artificial intelligence in bladder MRI is best positioned for quality auditing and acceleration rather than diagnosis. And don't assume every patient wants their biopsy result the moment it's signed — most, in one safety-net cohort, did not.

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 you have a paper you have been meaning to read, upload the PDF, or paste any link, at audioscholar dot C C. We will turn it into audio like this one, in any of thirty-one languages.

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