European Radiology

Beyond Morphology: T2 Mapping of Knee Cartilage in Sports-Active Adolescents 🦵🧲

How does repetitive sports activity affect developing knee cartilage before structural damage becomes visible? This study used quantitative T2 mapping to assess region-, zone-, and layer-specific cartilage patterns in sports-active adolescents. Higher T2 values were predominantly observed in the patellofemoral joint, lateral tibiofemoral compartment, and medial femoral condyle, with posterior and depth-dependent alterations associated with training exposure. 🏃

👉 https://link.springer.com/article/10.1007/s00330-026-12765-4 (Edis Çolak et al.)

"These findings highlight the potential of T2 mapping as an imaging biomarker of early cartilage changes, capturing sport-specific adaptations to repetitive mechanical loading before morphological abnormalities become apparent. Quantitative MRI may therefore offer valuable insights into cartilage health and help identify young athletes at risk of early cartilage degeneration. ✨" - Sonja Jankovic

Beyond the Plaque: Pericoronary Fat Radiomics for Predicting Cardiac Events 🫀📊

Pericoronary adipose tissue (PCAT) reflects local inflammation and may provide prognostic beyond conventional risk factors and assessment. This systematic review and meta-analysis evaluated radiomics on for predicting major adverse cardiac events (MACE). Radiomics models outperformed clinical or imaging models alone, while combined clinical, imaging, and models achieved the highest predictive performance. 🖥️

👉 https://link.springer.com/article/10.1007/s00330-026-12778-z (Seyedeh-Tarlan Mirzohreh et al.)

"These findings highlight the potential of PCAT radiomics as a quantitative imaging biomarker of coronary inflammation. Integrating radiomics with established clinical and imaging parameters may refine cardiovascular risk stratification, although standardized pipelines and prospective multicenter validation remain essential before clinical implementation. 📊" - Sonja Jankovic

CTEPH: where imaging and intervention make a difference 🫁

Chronic thromboembolic pulmonary hypertension is a rare but potentially treatable cause of pulmonary hypertension. Its management has evolved from a primarily surgical pathway to a multimodal strategy combining pulmonary endarterectomy, balloon pulmonary angioplasty, and targeted medical therapy.

For radiologists, CTEPH is a powerful example of imaging-driven therapeutic decision-making. Imaging must do more than confirm disease, it needs to map proximal and distal vascular involvement, assess perfusion and haemodynamic impact, and guide multidisciplinary treatment selection.

Emerging tools such as dual-energy CT, photon-counting CT, quantitative perfusion assessment, and computational fluid dynamics may help move CTEPH imaging beyond morphology toward functional phenotyping.

https://link.springer.com/article/10.1007/s00330-026-12781-4 (Laurent Godinas et al.)

ESR Bridges: chronic thromboembolic pulmonary hypertension: new developments in imaging and treatment—a multidisciplinary view - European Radiology Home European Radiology Article ESR Bridges: chronic thromboembolic pulmonary hypertension: new developments in imaging and treatment—a multidisciplinary view Commentary Open access Published: 05 August 2026 (2026) Cite this article You have full access to this open access article Download PDF Sav...

Desmoid-type fibromatosis can behave unpredictably, ranging from spontaneous regression to progressive enlargement. This makes active surveillance challenging and reliable imaging predictors highly desirable.

In this retrospective study, Nagata et al. evaluated 24 patients with desmoid-type fibromatosis managed with active surveillance, using qualitative MRI features, whole-tumor T2 ratio, ADC, and 3D volumetry.

The findings were surprising: none of the evaluated MRI feature, including T2 hyperintensity, band sign, tail sign, Yin–Yang sign, DCE pattern, T2 ratio, or ADC, was significantly associated with subsequent tumor growth, stability, or shrinkage.

"Desmoid fibromatosis is an elusive disease 🔮 and the search goes on to seek the best biomarker to assess for disease activity so treatment escalation can be planned ahead for this potentially debilitating illness" - Kevin Fung

https://link.springer.com/article/10.1007/s00330-026-12777-0 (Shuji Nagata et al.)

Can MRI replace CT for pediatric skull base bone assessment? 💀 👶🏻

CT remains the reference standard for evaluating skull base bone lesions, but radiation exposure is an important concern in children.

In this prospective study, Yin et al. evaluated the FRACTURE MRI sequence as a radiation-free alternative to CT in 40 pediatric patients with suspected skull base lesions.

FRACTURE provided CT-comparable bony delineation, sharper bony edge definition, and superior soft-tissue visualization. Diagnostic accuracy was equivalent to CT, with high consistency in treatment planning.

Importantly, this allows assessment of both bone and adjacent soft tissues within a single MRI-based protocol, potentially reducing the need for separate CT imaging

https://link.springer.com/article/10.1007/s00330-026-12774-3 (Guangheng Yin et al.)

Radiology AI: are we focusing on real-world impact? 🤖🌍

AI tools in radiology are often evaluated under controlled conditions, but their true value depends on how they perform in routine clinical practice.

In this commentary, Plasencia-Martínez discusses a real-world evaluation of a commercial AI tool for detecting intracranial hemorrhage on emergency CT. The study is important not only because it was prospective, used consecutive cases, different CT scanners, with a large sample size, but also because it reported limited clinical benefit instead of another positive headline.

Negative or neutral AI studies matter. They help counter the perception that AI tools are inherently reliable and provide the evidence needed to decide which tools genuinely improve care, and which do not.

"What do you think? Can AI survive out of its sandbox in the wild west?" - Kevin Fung

https://link.springer.com/article/10.1007/s00330-026-12772-5 (Juana María Plasencia-Martínez)

Demonstrating the real-world impact of an artificial intelligence tool in radiology: are we focusing our efforts in the proper direction? - European Radiology Home European Radiology Article Demonstrating the real-world impact of an artificial intelligence tool in radiology: are we focusing our efforts in the proper direction? Commentary Published: 21 July 2026 (2026) Cite this article Save article View saved research European Radiology Aims and scope Sub...

Can general-purpose LLMs improve radiology reporting efficiency? 🤖📜

Speech recognition has become routine in radiology reporting, but transcription errors, spelling mistakes, and editing burden remain familiar challenges.

In this prospective multicenter study, de Margerie-Mellon et al. compared conventional speech recognition with a general-purpose LLM with built-in speech recognition during routine CT and MRI reporting.

Overall, LLM-assisted reporting was faster and produced fewer grammar, spelling, and transcription errors. However, the benefit was not uniform: only 3 of 5 radiologists achieved shorter total reporting times.

Importantly, new error patterns also emerged, including prompt non-compliance, rewording without meaning change, and occasional confabulations.

"General-purpose LLMs may improve reporting efficiency, but they are not simply a plug-and-play replacement for speech recognition. Human in the loop oversight is still essential." - Kevin Fung

https://link.springer.com/article/10.1007/s00330-026-12524-5 (Constance de Margerie-Mellon et al.)

Can the 20-minute hepatobiliary phase reveal where PSC is heading next? ⏱️🩻

This retrospective study of 73 patients evaluated the prognostic value of gadoxetate-enhanced MRI, finding that the relative enhancement of proximal extrahepatic bile ducts (REPD) at 20 minutes was a significant predictor of adverse events within two years, while the ANALI score demonstrated an 87.5% sensitivity in predicting the need for liver transplantation.

The accompanying editorial highlights the important translational challenge: quantitative ductal signal measurements are promising but technically demanding, while simpler qualitative markers such as potential functional stricture may prove more reproducible and practical.

Highly recommend reading this paper and the insightful commentary for a balanced view of how functional MRI could improve PSC prognostication—and what is still needed to make it work in everyday practice.

"This highlights the shift toward using hepatobiliary contrast not just for static anatomical staging, but as a dynamic, functional biomarker to refine risk stratification for chronic liver disease." - Amit Gupta

https://link.springer.com/article/10.1007/s00330-026-12710-5 (Kumaresan Sandrasegaran et al.)

Can one 17-second breath-hold turn lung MRI into a credible radiation-free nodule screening tool? 🫁🧲

In this prospective study of 97 adults, an AI-powered compressed-sensing 3-T MRI sequence acquired in a single breath-hold was compared with photon-counting CT, which identified 118 pulmonary nodules. MRI achieved 83.1% sensitivity and 87.3% accuracy for solid and part-solid nodules overall, rising to 98.1% sensitivity and 97.7% accuracy for solid nodules ≥ 4 mm, with complete agreement for Lung-RADS 4A and 4B lesions; however, performance remained limited for very small and calcified nodules.

A fascinating read for anyone exploring radiation-free thoracic imaging - because it shows how AI-accelerated MRI is moving from technical possibility toward clinically meaningful nodule detection.

"The ability to accurately detect and size clinically relevant lung nodules without ionizing radiation marks a significant leap toward safer, high-throughput longitudinal screening protocols." - Amit Gupta

https://link.springer.com/article/10.1007/s00330-026-12738-7 (Anna Palmisano et al.)

Can an LLM catch a report error that any radiologist would spot in seconds? Not reliably - yet 🤖🩻⚠️

This study tested GPT-4.1 and Llama 3.3 70B on 1,024 error-containing reports, generated from 256 CT, MRI, and radiography reports, and found overall error-detection accuracies of 66.5% and 54.0%, respectively; when both detection and correct classification were required, performance fell to 48.2% and 36.0%. Both zero-shot models performed better on inappropriate recommendations but struggled with reasoning-dependent mistakes: for physiologically impossible errors on MRI, two-stage success dropped to just 8.7% for GPT-4.1 and 1.9% for Llama 3.3 70B.

A fascinating and cautionary read for anyone exploring LLM-based report quality assurance - because fluent language should never be mistaken for reliable clinical reasoning.

"LLMs may eventually become useful reporting safety nets, but this study makes it clear that current general-purpose, zero-shot models cannot yet be trusted as autonomous error checkers 🧠🛡️✨. " - Amit Gupta

https://link.springer.com/article/10.1007/s00330-026-12697-z (Tugba Akinci D’Antonoli et al.)