European Radiology

New Zealand White rabbits are widely used in interventional radiology research due to their suitability for human-sized treatment and imaging equipment, offering high translational potential. This study aimed to define selection criteria for rabbits by correlating body weight and age with abdominal organ and vessel dimensions measured on cross-sectional imaging

https://link.springer.com/article/10.1186/s41747-026-00708-z (Selma Saclier et al.)

What if the ultimate fix for chaotic, unreliable DICOM metadata has been hiding in plain sight on the standard CT scout view all along? 🤖🩻🧭

This retrospective study introduces RAPID, a deep learning framework that combines CT topogram analysis with DICOM spatial geometry to label the anatomical coverage of CT series directly from image data, rather than relying on inconsistent metadata. Across 65,250 patients and 104,009 topograms, anatomical classification achieved F1 scores of 0.92 internally and 0.97 externally, while external F1 scores remained 0.93 for body-region detection and 0.91 for landmark detection; importantly, radiologists rated more than 98% of reviewed predictions as good or excellent.

A very practical read for anyone working with large imaging datasets or clinical AI pipelines, because reliable identification of what anatomy a CT series actually contains is a deceptively important prerequisite for automated routing, cohort selection, and downstream analysis.

https://link.springer.com/article/10.1007/s00330-026-12830-y (Yutong Wen et al.)

"Some of the most enabling radiology AI may operate quietly in the background. Moving from fragile text-based metadata toward image-derived anatomical understanding could make imaging datasets cleaner, workflows more scalable, and every downstream AI application more reliable." - Amit Gupta

Given the distinct intratumoral heterogeneity of glioblastoma (GB) and solitary brain metastasis (SBM), habitat-based radiomics derived from neurite orientation dispersion and density imaging (NODDI) may offer enhanced diagnostic value. This study aimed to evaluate NODDI habitat analysis in distinguishing GB from SBM.

https://link.springer.com/article/10.1186/s41747-026-00717-y (Wenzheng Luo et al.)

Can photon-counting CT make pulmonary angiography lighter on both radiation and iodine, while maintaining image quality? 🫁⚡🩻

This systematic review and meta-analysis of 12 studies involving 950 patients found that PCCT-based CTPA was associated with a 2.85 mGy lower CTDIvol (approximately 30% of the average EID-CT dose) and 5.62 gI lower iodine load, while maintaining comparable SNR and CNR. However, these benefits lost statistical significance under conservative statistical adjustments designed to account for extreme protocol variations across the available studies.

A valuable read for anyone following photon-counting CT, because its potential lies not only in better images, but in making CTPA safer and more efficient for patients who may benefit most from dose and contrast reduction.

https://link.springer.com/article/10.1007/s00330-026-12829-5 (Shingo Kato et al.)

"An exciting secondary highlight from this analysis is that PCCT-generated spectral iodine maps demonstrated a pooled sensitivity of 92.3% and specificity of 83.4% for chronic thromboembolic disease, highlighting a powerful emerging tool for simultaneous structural and functional pulmonary evaluation without the need for additional radiation 🎯📉✨. " - Amit Gupta

Can an open-weight AI model successfully unlock structured clinical data from free-text non-English brain MRI reports without compromising patient privacy? 🤖🧠📄

This retrospective study evaluated the open-weight LLM LLaMA 3.1 across 947 Dutch brain MRI reports from a memory clinic to automatically extract 30 structured variables spanning visual rating scales, vascular findings, counts, and free-text locations. The model achieved near-human accuracy for standardized visual rating scales such as Fazekas (0.94) and medial temporal lobe atrophy (up to 0.96), with similarity-based few-shot prompting significantly boosting the extraction of complex numerical lesion counts up to 0.92.

Highly recommend reading this paper if you are looking to deploy open-source language models for automated clinical data curation.

https://link.springer.com/article/10.1007/s00330-026-12821-z (Kaouther Mouheb et al.)

"This demonstrates that privacy-preserving, locally deployable models can effectively bridge the gap between radiologist reporting freedom and the structured data needs of large-scale neuroimaging registries. 🔓📊✨. " - Amit Gupta

What if the traditional metrics tracking radiology department speed and compliance are completely missing the hidden wave of scans that generate clinical ambiguity instead of answers? 🎯🩻

This thought-provoking commentary examines how soaring cross-sectional imaging volumes, now surpassing 370 annual examinations per 1,000 inhabitants in several industrialized nations, threaten to fuel low-value cascades, incidentalomas, and overtreatment. The author advocates for systematically tracking "diagnostic yield" as an essential quality indicator alongside turnaround times, directly assessing whether examinations produce definitive, actionable clinical answers.

A timely read for radiologists and referrers alike, because imaging quality should perhaps be measured not only by how well and how quickly we scan, but by how often the examination meaningfully informs patient care.

https://link.springer.com/article/10.1007/s00330-026-12813-z (Benoît Gallix)

"Diagnostic yield offers a compelling way to bring value into everyday imaging quality assessment; as feedback to help us achieve what matters most: the right test, for the right patient, at the right time 🧭🤝✨. " - Amit Gupta

Diagnostic yield: a missing quality metric in medical imaging - European Radiology Home European Radiology Article Diagnostic yield: a missing quality metric in medical imaging Commentary Published: 18 August 2026 (2026) Cite this article Save article View saved research European Radiology Aims and scope Submit manuscript Benoît Gallix ORCID: orcid.org/0000-0003-0050-86081,2,3 66...

Acute heart failure is a common but underrecognised cause of dyspnea. Chest CT can accurately assess pulmonary congestion, but radiologist reporting capacity may limit clinical utility. The authors hypothesise that an AI model could automatically detect imaging signs of acute heart failure

https://link.springer.com/article/10.1186/s41747-026-00718-x (Kristina Cecilia Miger et al.)

Smarter CT Data Processing: Can AI Make Series Labelling More Reliable? 🤖

Accurate anatomical labelling of CT series is fundamental for efficient clinical workflows and large-scale imaging analysis, yet conventional approaches often rely on inconsistent DICOM metadata. This study introduces RAPID, a deep learning framework that analyzes routinely acquired CT topograms to automatically identify body regions and anatomical landmarks, subsequently mapping this information to individual CT series. The models demonstrated high performance across both internal and external datasets. 🧠📊

👉 https://link.springer.com/article/10.1007/s00330-026-12830-y (Yutong Wen et al.)

"These findings show how information already available in routine imaging, the CT topogram, can be transformed into a powerful tool for anatomy-aware data processing. ✨ By reducing dependence on manual review and unreliable textual metadata, automated anatomical labelling could improve data consistency, scalability, study navigation, and downstream AI workflows in both clinical practice and imaging research. 🚀" - Sonja Jankovic

AI for Longitudinal Pulmonary Nodule Matching: Toward More Automated Lung Cancer Screening 🤖🫁

Accurate longitudinal matching of pulmonary nodules is essential for assessing nodule growth and volume doubling time (VDT) in lung cancer screening. This study evaluated a fully automated AI system on follow-up low-dose CT, achieving successful matching in 83.5% of persisting nodules, increasing to 91.8% in patients with a single baseline nodule. Importantly, only 1.5% of persisting nodules required manual correction, highlighting the potential for substantial workload reduction. 🩻📊

👉 https://link.springer.com/article/10.1007/s00330-026-12825-9 (Beibei Jiang et al.)

"These findings bring AI one step closer to a fully automated longitudinal lung cancer screening workflow. Reliable nodule matching could streamline VDT assessment and reduce repetitive manual tracking, allowing radiologists to focus their expertise where it matters most. Further validation in diverse screening populations will be important, particularly in patients with a high nodule burden. ✨" - Sonja Jankovic

🧡 The European Society of Radiology and the ESR Journals are excited to announce the publication of its debut series of articles in European Radiology Abdomen. The journal, established as a joint venture between the ESR, European Society of Gastrointestinal and Abdominal Radiology (ESGAR), and European Society of Urogenital Radiology (ESUR), has published its first papers, including the inaugural editorial written by the editors of the journal.

You can read it, along with all future articles at the Springer Nature Link ➡️

https://buff.ly/zOyVs5t