Weekly Radiology Review

Key Finding: LLM-generated lumbar spine MRI reports were indistinguishable from human reports for non-specialized readers, suggesting potential to support radiological workflow.

Key Finding: GPT-4o optimized with detailed prompting selected optimal abdominal/pelvic CT protocols more frequently than human radiologists, without requiring fine-tuning.

Key Finding: Merlin, a 3D vision-language model trained on abdominal CTs, outperformed existing models in diagnostic, prognostic, and quality-related tasks with high cross-institutional generalization.

Key Finding: MARS-MRI showed sensitivity of 0.88 and specificity of 0.95 for periprosthetic joint infection diagnosis (AUC 0.94), proving valuable for guiding antimicrobial stewardship.

Key Finding: Whole-body MRI proved a promising modality for staging and follow-up of primary musculoskeletal tumours, with potential to replace multimodal protocols in sarcoma imaging.

Key Finding: A novel hybrid neural network (DOGHNN) achieved accurate white matter hyperintensity segmentation in clinical MRI, with strong potential for automating neuroimaging workflows.

Key Finding: CT and MRI showed substantial to near-perfect agreement in grading brain atrophy and white matter changes in ischemic stroke, validating CT for brain frailty assessment in clinical practice.

Key Finding: AI offered higher specificity while visual scales showed greater sensitivity for MRI-based dementia evaluation; lobar microbleeds were more frequent in neurodegenerative cases.

Key Finding: 18F-FDG PET/CT-based radiomic models preoperatively predict spread through air spaces (STAS) in lung cancer, enabling noninvasive risk stratification for surgical planning.

Key Finding: Baseline total tumor SUVmean on PSMA PET/CT is an independent prognostic biomarker in metastatic castration-resistant prostate cancer treated with 177Lu-PSMA-617.

Key Finding: FDG PET and PSMA PET provide distinct yet complementary prognostic insights in advanced hormone-sensitive prostate cancer, enhancing combined prognostic accuracy.

Key Finding: Updated ESUR guidelines confirm the safety of non-ionic iodine-based GI contrast agents and provide guidance for managing barium-related bowel perforation risks.

Key Finding: Pharmacovigilance analysis reveals heterogeneous respiratory adverse event profiles among four non-ionic iodinated contrast media, identifying reactions not covered in prescribing information.

Key Finding: AI consistently enabled dose reductions of 30%–95% across CT applications while maintaining diagnostic adequacy, with strongest evidence in chest screening, oncology, vascular imaging, and pediatrics.

Key Finding: Systematic review highlights the need for international standardization of clinical indication-based Diagnostic Reference Levels (NDRLci) to reduce dose discrepancies across countries.