Ultrasound fat fraction graded steatosis against biopsy in 255 patients, whatever the liver disease

The edition in brief

Ultrasound-derived fat fraction was tested prospectively in 255 patients with chronic liver disease of mixed aetiology, all with liver biopsy as reference; 57.25% had no steatosis. Area under the curve was 0.914 (95% CI 0.873 to 0.946) for steatosis of grade S1 or above, 0.950 (0.916 to 0.974) for S2 or above and 0.916 (0.875 to 0.947) for S3, with optimal cutoffs of 7%, 11% and 18%. Performance stayed above 0.9 in patients with necroinflammation, with advanced fibrosis, with a skin-to-capsule distance of 1.9 cm or more, and after excluding isolated metabolic dysfunction-associated steatotic liver disease. In 436 patients with Ménière's disease, 872 ears, GPT-5 graded endolymphatic hydrops on delayed gadolinium-enhanced 3D-FLAIR MRI at 25% accuracy for cochlear and 39% for vestibular grading under basic prompting — worse than chance for a graded scale in practice. With structured image descriptions, clinical context and worked examples supplied, accuracy rose to 87% and 80%, and junior readers working with the model improved from 36% to 59% for cochlear grading. The model was not reading the images so much as reasoning over descriptions of them. A meta-analysis of three studies and 97 patients found a pooled 18F-FDG PET positivity rate of 94% (73 to 99, I² 0%) in relapsing polychondritis, with pooled SUVmax 4.0, asymptomatic cartilage involvement in more than a quarter, and better yield at inaccessible sites such as the peripheral airways. And the ACR sets out how it intends to build paediatric-specific incidental finding guidance, with an interim strategy for the many findings where adult follow-up algorithms do not apply to a child.