Radiology
Ketamine with midazolam gave less pain and fewer desaturations than fentanyl in radiologist-led sedation
A randomised trial of 264 CT- and ultrasound-guided procedures favoured ketamine, at the cost of more hallucinations. An AI model caught radiograph-occult femoral neck fractures that readers missed, and routine head CT doses varied up to 3.4-fold between common protocols.
An enhancement curve drawn by hand made O-RADS MRI more specific for adnexal cancer
A semi-quantitative intensity curve raised O-RADS MRI specificity from 0.84 to 0.92; a simple cine-derived coupling ratio flagged heart failure risk in thalassaemia; and the bone tumour reporting systems held up in children.
AI risk scores could spare about one in five PI-RADS 3 biopsies while keeping sensitivity near 98%
A multicentre prostate MRI study positions AI as a triage aid for equivocal lesions; the ERS asks for HRCT screening of every adult with CVID; FDG PET-CT beats CT for response after oesophagogastric chemotherapy; and PSA density has a ceiling in healthy men.
AI reading of non-contrast CT flags more large vessel occlusions than human readers — as a prompt, not a verdict
A network meta-analysis puts AI sensitivity for LVO at 78% against about 60% for readers, PI-QUAL makes prostate MRI quality something to report, and a biopsy trial finds the diagnosis, not the needle, drives quality of life.
One in seven patients given a follow-up imaging recommendation had a diagnostic error, mostly through delay; photon-counting CT rivals breast MRI for staging
A quality study of abdominal imaging finds most diagnostic errors happen after the report is signed, photon-counting CT matches MRI on breast lesion extent and beats it on nodes, deep learning finds cerebral microbleeds accurately in small studies, and six societies call for mandatory enhanced radiation protection in fluoroscopy labs.
Ultra-low-dose CT at 1.7 mSv replaced most ultrasound and KUB for renal stone surveillance
A quality programme moved stone follow-up to 1.7 mSv CT; an aneurysm AI run silently alongside radiologists found 55 extra aneurysms in 3,856 CTAs; commercial nodule AI cut chest CT reporting time by about 15%; and ancillary features sharpened Carotid Plaque-RADS.
Deauville 5a held up in two German Hodgkin trials: without new lesions it still marked a small, high-risk group at interim PET
A trial-population validation of the split Deauville 5 score, a post hoc BRIDGE-TNK analysis tying tenecteplase benefit to lower ASPECTS, a reader study of LLM assistance by expertise, and the MRI signs that separate leiomyosarcoma from fibroids.
Three signs on liver MRI that mark a small HCC as likely to come back
Hypovascular component, intratumoral artery and incomplete capsule on Gd-EOB-DTPA MRI identify the macrotrabecular-massive subtype and a 2.5-fold recurrence risk after resection. Plus what photon-counting CT changes in the head and neck, and an FDG-PET signal in relapsing polychondritis.
The fractures radiographs hide are the ones a model found
A deep learning model reached 94.7% sensitivity for radiograph-negative femoral neck fractures where emergency physicians reached 68.8%; plus updated appropriateness criteria for myelopathy, a threefold spread in routine head CT dose, and a sedation regimen worth switching to.
BI-RADS v2025 puts nonmass lesions into the breast ultrasound lexicon — and they need tissue
A lexicon change that alters what you report and what happens next, alongside the first full ASNC stress testing rewrite since 2016, a working-group verdict on prostate MRI AI, and a rubric for the plain-language reports your department is probably already sending out.
PET-CT earns its place after chemotherapy, but not for staging
In oesophagogastric adenocarcinoma, PET-CT and contrast CT predicted pathological stage equally badly — but PET found every metastasis and CT found half. Plus Bone-RADS in children, what GPT-5 can and cannot do with a 3D-FLAIR, and a sex signal in CT-FFR worth watching but not acting on.
Adding an enhancement curve to O-RADS MRI halves the benign masses sent to oncology surgeons
A 341-mass series raises O-RADS specificity from 0.84 to 0.92 with a semi-quantitative curve; standalone AI matched 83.5% of screening nodules across scans; and PI-QUAL gives prostate MRI quality a score that can be audited.
Observing and rescoring equivocal appendicitis sorted patients for CT better than a single score
A secondary analysis from the DIAMOND trial closes the edition. Before it: stress perfusion MRI giving answers to patients with angina and open arteries, right ventricular coupling in thalassaemia, photon-counting CT in breast cancer staging, and a call to make enhanced radiation protection mandatory.
AI found 55 aneurysms radiologists missed on 3,856 CT angiograms, and ultra-low-dose CT displaced ultrasound for stone follow-up
A prospective shadow-mode study measures what an aneurysm detection tool adds, and where it does not pay; nodule AI cuts chest CT reporting time by 15%; and CT attenuation beats MRI scoring for opportunistic osteoporosis detection.
The model helped the residents most, and misled them most
A reader study in thoracic imaging finds the benefit of a language model runs through the expertise of the person using it; a deep learning segmentation beats LAA-950 for emphysema; and ancillary features lift Plaque-RADS where stenosis alone decides nothing.
The model ranked the applicants, and pushed the same groups down every time
Seven model configurations produced 140 rank order lists from application data alone. None came close to the committee's, and all of them placed women, international and non-MD applicants lower than the humans did. Plus what a skull-base report has to contain, where robotic CT guidance earns its place, and a national registry using language models to watch imaging algorithms.
A threefold dose spread on the same head CT, and an AI that finds the fracture you cannot see
Clustering 904,209 head CT scans shows routine protocols differing up to 3.4-fold in dose; ketamine beats fentanyl for procedural sedation in interventional radiology; and a deep learning model reaches 94.7% sensitivity for radiograph-occult femoral neck fractures.
Same yield, a quarter of the chest drains: robotic bronchoscopy against CT-guided lung biopsy
The first review restricted to direct comparisons finds diagnostic yield indistinguishable (RR 0.99) and pneumothorax needing a drain or admission cut by about three-quarters — on five retrospective series from one health system, which is the catch. Plus the 2026 ASNC stress testing guideline, a safety rubric for AI-written patient reports, and non-mass lesions entering the ultrasound lexicon.
Ultrasound fat fraction graded steatosis against biopsy in 255 patients, whatever the liver disease
A prospective multicentre study puts ultrasound-derived fat fraction at an AUC of 0.914 for any steatosis and 0.950 for moderate or worse, holding up across aetiology, fibrosis stage and body habitus. Plus what GPT-5 can and cannot do with a 3D-FLAIR study, FDG PET in relapsing polychondritis, and the paediatric incidental finding problem.
Drawing the curve, not just looking at it, sharpens O-RADS MRI
In 341 surgically resected adnexal masses, adding a semi-quantitative enhancement curve to visual assessment raised specificity from 0.84 to 0.92 and positive predictive value from 0.77 to 0.90. Plus what a CT-FFR subgroup does and does not show about women, how to actually use a breast risk model, and PI-QUAL as a routine step rather than a research tool.
Perfusion selection extends thrombolysis to 24 hours, and dosing children's contrast by lean weight cuts the volume without losing enhancement
Six randomised trials support intravenous thrombolysis out to 24 hours in perfusion-selected stroke at the cost of more symptomatic haemorrhage, an AI system matched 83.5% of persisting nodules across screening rounds, and adding contrast sequences beats endorectal ultrasound in early rectal cancer.
Saline in the biopsy tract cut pneumothorax by half and chest tubes by nearly 80%
A meta-analysis of seven studies found instilling sterile saline into the tract after CT-guided lung biopsy reduced pneumothorax risk to less than half, the ACR has updated its myelopathy appropriateness criteria, and commercial AI cut pulmonary nodule reporting time by 14.6% while slowing emergency department reporting.
Commercial AI shaved 15% off chest CT reporting time across 39,000 scans
A real-world before-and-after study puts a number on what nodule AI does to throughput, saline tract sealing more than halves pneumothorax after CT-guided lung biopsy, and photon-counting CT matches MRI for breast staging while missing nearly half the microcalcifications.
Commercial AI cut pulmonary nodule reporting time by 15% across 39,000 chest CTs
A real-world before-and-after study puts a number on what AI does to reporting time, saline tract sealing more than halves pneumothorax after CT-guided lung biopsy, and photon-counting CT matches MRI for breast staging while missing nearly half the microcalcifications.