BI-RADS v2025 puts nonmass lesions into the breast ultrasound lexicon — and they need tissue

The edition in brief

The BI-RADS v2025 Manual formally brings nonmass lesions into the breast ultrasound lexicon, with descriptors organised by distribution — segmental, linear, focal or regional — plus echo pattern, posterior features and associated findings. A pictorial review sets out pathologically confirmed benign and malignant examples, notes the overall high malignancy rate that warrants tissue diagnosis, and details why these lesions are missed: no discrete margin or shape, the gravitational effect of patient positioning, transducer orientation during real-time scanning, and the operator dependence of handheld acquisition. The American Society of Nuclear Cardiology has rewritten its stress testing guideline for the first time since 2016, covering patient selection and preparation, stress modalities and protocols, safety, and interpretation and reporting, for SPECT and PET myocardial perfusion imaging. The ESUR Prostate MRI Working Group reports that deep learning tools for prostate cancer detection have high technical accuracy but a wide gap to clinical readiness, recommending a human-in-the-loop model to limit automation bias and liability, prospective multi-vendor validation, and post-market surveillance for algorithmic drift. A diagnostic meta-analysis of 20 studies found apparent diffusion coefficient grading of renal cell carcinoma had pooled sensitivity 0.84 and specificity 0.57, summary AUC 0.71, with radiomics and deep learning models reaching sensitivity 0.79, specificity 0.86 and AUC 0.90 across only five studies. And a prospectively developed five-attribute rubric — clarity, content, certainty, tone, verbosity — offers a quality gate before AI-generated patient-friendly reports are released, with radiologist interreader agreement of 0.65 and an AI-applied version agreeing with reference distribution decisions 88.1 per cent of the time.