AI reading of non-contrast CT flags more large vessel occlusions than human readers — as a prompt, not a verdict

The edition in brief

A Bayesian network meta-analysis of 10 retrospective validation studies (3,632 patients) found imaging AI detected anterior circulation large vessel occlusion on non-contrast CT with sensitivity 0.78 against 0.62 for expert and 0.60 for non-expert readers, with similar specificity around 0.86–0.88. Certainty was low and no study measured time to reperfusion or outcome, so AI belongs as an escalation prompt towards CT angiography or tele-stroke review, not as a rule-out. In the DEPROMP trial of 230 biopsy-naive men, combined systematic, MRI-targeted and PSMA PET-targeted transrectal biopsy had low morbidity, and worse quality of life at six months was attributed to the cancer diagnosis, not the biopsy. In the UK Lung Cancer Screening trial, an AI system matched 83.5% of persisting nodules across scans automatically, and only 1.5% of real nodules needed manual correction. An AJR expert panel sets out a practical approach to breast cancer risk assessment for screening. The practice-changer: an AJR special series review sets out how to put PI-QUAL image-quality scoring into routine prostate MRI, because poor-quality scans increase equivocal PI-RADS reports and weaken the pathway that depends on them.