A brain aneurysm is a weakened area in the wall of a blood vessel in the brain that can bulge outward. Unruptured aneurysms are relatively common, though many produce no symptoms and are discovered incidentally.

If an aneurysm ruptures, it can cause bleeding around the brain called a subarachnoid hemorrhage. This is a medical emergency, making accurate detection important for guiding timely treatment and reducing the risk of potentially fatal complications.

Radiologists can often identify brain aneurysms through CT imaging, particularly CT angiography, or CTA. These examinations help clinicians assess an aneurysm’s location, size, shape, and relationship to nearby blood vessels when determining appropriate clinical management.

A new prospective study suggests that a Food and Drug Administration cleared artificial intelligence algorithm could help radiologists find more aneurysms on CT scans. The findings were published in the Journal of the American College of Radiology.

Researchers studied 3,856 CTA examinations performed across the Northwell Health system. They evaluated aiOS, an algorithm developed by Aidoc to detect intracranial aneurysms.

The AI analyzed scans alongside routine clinical care, but radiologists could not see its findings during their initial interpretations. This design allowed researchers to evaluate the technology in an everyday clinical environment and compare its results with physician assessments.

Radiologists and the AI agreed in more than 96% of examinations. The algorithm identified 55 aneurysms that initial radiologists missed, producing a 39% relative enhanced detection rate.

The AI demonstrated greater sensitivity than radiologists alone, detecting 84.6% of aneurysms compared with 71.8% for radiologists. That result means the algorithm found more aneurysms that were genuinely present.

Radiologists, however, were more likely to be correct when reporting that an aneurysm was present. Their positive predictive value reached 92.7%, compared with 78.2% for the AI.

Both approaches showed similar strengths in ruling out aneurysms when none were present and avoiding false alarms. The findings therefore suggested that the algorithm and radiologists offered different, potentially complementary capabilities.

Shlomit Stein, MD, FACR, is Professor of Radiology at the Zucker School of Medicine at Hofstra/Northwell and Director of Artificial Intelligence in the Department of Radiology at Northwell Health. Stein was the study’s lead author.

“AI was able to identify mostly small aneurysms missed by radiologists. Despite their small size, they may nevertheless be clinically important since risk is not only a function of aneurysm size, but also shape, location, and other patient-related risk factors,” Stein told Medical News Today.

The AI found 55 aneurysms that radiologists missed, while radiologists detected 30 that the algorithm did not. Among 101 findings identified only by the AI, 46 were ultimately judged to be false positives.

Researchers found that additional true aneurysm detections outweighed the false positive findings and improved overall detection performance. At the same time, radiologists identified important aneurysms that the algorithm missed, reinforcing the potential value of combining the two approaches.

The algorithm’s usefulness also varied according to the care setting. Its strongest performance occurred in inpatient examinations, where it identified 18 additional aneurysms while generating seven false positive alerts, and it also performed favorably in the emergency department.

The benefit was more limited in outpatient care. There, the AI detected four additional aneurysms but generated more false positive findings than true positive findings.

Researchers suggested that differences in patient populations and examination complexity could help explain the variation. Inpatient and emergency settings may include more complex cases, providing more opportunities for AI to act as an additional detection tool.

The study also showed why medical AI must be assessed under routine clinical conditions. Performance can differ when an algorithm encounters the range of patients, imaging equipment, clinical indications, and workflows present in everyday healthcare.

Continued monitoring after AI systems enter clinical practice remains important, according to the researchers. Their results indicate that AI could provide another layer of review for brain CTA examinations while leaving final interpretation and clinical decisions to physicians.

The study did not establish that AI assisted detection improves patient outcomes. More research is needed to determine whether finding additional aneurysms ultimately reduces ruptures or other complications.

“Aneurysms carry the risk of rupture and potentially catastrophic brain hemorrhage. Some of these detected aneurysms will therefore require surveillance or preventative intervention. The first step is aneurysm detection, which we have shown can be aided by the use of AI,” Stein concluded.