An AI read found 55 aneurysms radiologists missed on 3,856 CT angiograms

The edition in brief

In a prospective shadow-mode study of 3,856 consecutive brain CT angiograms, a commercial FDA-cleared algorithm was more sensitive than the reporting radiologists for intracranial aneurysm (0.85 vs 0.72) with similar specificity. It found 55 aneurysms the radiologists missed, against 46 false alerts, and radiologists found 30 the algorithm missed; most AI-only finds were under 3 mm. The yield was best in inpatients and poorest in outpatients. A meta-analysis of 16 studies (1,999 patients) found pooled sensitivity for histologically proven pelvic nodal metastasis of about 55% for PSMA PET/CT and 33% for MRI, with specificity above 90% for both, so a negative scan does not exclude microscopic nodal disease. In 418 resected IPMNs, current high-risk stigmata had 83% sensitivity but only 67% specificity for malignancy; raising the enhancing mural nodule and duct cut-offs improved specificity at the cost of sensitivity. A US multisite system that required high-priority diagnostic breast imaging within three weeks raised on-time completion from 57% to 65% and cut the median wait from 18 to 14 days.