Tejal Somvanshi
A real-world study of 3,856 brain scans found that an artificial intelligence system can help radiologists detect more aneurysms—but with an important catch: the AI flags more potential problems, including cases that turn out to be false alarms.
According to a study published in the Journal of the American College of Radiology in mid-September, researchers at Northwell Health tested an FDA-cleared AI tool from a company called Aidoc against radiologists reading CT angiography scans, which are used to look for aneurysms inside the brain.
The core finding was straightforward: when radiologists used the AI as an assistive tool, they missed fewer aneurysms. The algorithm identified 55 aneurysms that the radiologists alone did not catch in their initial readings—a 39 percent relative increase in detection. In raw numbers, radiologists found 71.8 percent of aneurysms present in the scans; with AI assistance, that rose to 84.6 percent.
But here’s the trade-off. Radiologists were more accurate when they identified an aneurysm on their own, getting it right 92.7 percent of the time. The AI was correct only 78.2 percent of the time when it flagged a potential aneurysm. That means the AI generates more false positives—cases flagged as aneurysms that aren’t actually aneurysms.
Both radiologists and the AI were equally strong at one critical task: ruling out aneurysms when none were present and avoiding false alarms that might trigger unnecessary patient anxiety or follow-up procedures.
The practical implication is that AI works best as a second opinion rather than a replacement. The algorithm catches things radiologists miss on first glance, but a radiologist’s judgment remains essential for confirming whether a finding is genuinely an aneurysm or an artifact. This is especially important because false alarms carry real consequences: a patient might spend months worrying about a brain aneurysm they don’t have or undergo unnecessary follow-up imaging.
For patients, the study doesn’t suggest they need to seek out AI-assisted screening independently. Aneurysms are detected when someone gets a CT scan for another reason—usually because of symptoms like a sudden severe headache, or as an incidental finding during imaging for a different medical issue. The AI wouldn’t change when or why you’d get scanned. What it might do is make it slightly more likely that a genuine small aneurysm gets caught by a radiologist using AI-assisted interpretation.
The study involved routine clinical scans from a large hospital network rather than a controlled research setting, which means the results reflect what actually happens in everyday medical practice. That matters because lab studies sometimes produce results that don’t hold up in the real world.
Brain aneurysms are relatively uncommon, and most never rupture. Ruptured aneurysms account for a small but significant share of all strokes. Detecting aneurysms that have a higher risk of rupture allows doctors to treat them before they cause a bleed.
The findings add to growing evidence that AI can serve as a useful tool in radiology, improving detection of abnormalities without replacing the expertise radiologists bring to image interpretation. The real-world test confirms that the technology works outside controlled trial conditions, which is an important step toward broader clinical adoption.
