AI found 55 aneurysms radiologists missed on 3,856 CT angiograms, and ultra-low-dose CT displaced ultrasound for stone follow-up

The edition in brief

Five findings for radiologists and the clinicians who request imaging. In a prospective shadow-mode study of 3,856 consecutive brain CT angiograms, an FDA-cleared AI tool detected aneurysms with sensitivity 0.846 against 0.718 for radiologists, at similar specificity, finding 55 true aneurysms radiologists missed; 56% of those were under 3 mm, and the yield was highest in inpatients and lowest in outpatients. In 39,323 chest CTs at a Dutch tertiary centre, commercial nodule AI was associated with median reporting time falling from 21.3 to 18.2 minutes, though emergency examinations took 7% longer. A retrospective study of 264 cirrhotic patients with poor ultrasound visualisation found non-contrast abbreviated MRI gave adequate views in over 91%, and 17 of 18 cancers detected were early stage. A meta-analysis of 42 studies found CT Hounsfield units outperformed MRI vertebral bone quality scores for detecting osteoporosis (AUC 0.86 vs 0.79), with a pooled threshold of 104.6 HU. A quality improvement project introducing tin-filter ultra-low-dose CT for renal stone surveillance cut dose to 1.7 mSv and ultrasound use by 71%.