Built upon an algorithm called HeadXNet, the tool highlights areas in the brain that are likely to contain an aneurysm. During testing, it helped doctors correctly identify six more aneurysms per 100 scans that displayed the condition. According to the team, the AI also improved consensus among the clinicians that took part in the trial. The research is published in Jama Network Open.
"Search for an aneurysm is one of the most labour-intensive and critical tasks radiologists undertake," said Kristen Yeom, associate professor of radiology at Stanford and co-senior author of the paper. "Given inherent challenges of complex neurovascular anatomy and potential fatal outcome of a missed aneurysm, it prompted me to apply advances in computer science and vision to neuroimaging."
To train the algorithm, Yeom worked with co-author Allison Park, a Stanford graduate student in statistics, and Christopher Chute, a graduate student in computer science. Together they outlined clinically significant aneurysms detectable on 611 computerised tomography (CT) angiogram head scans.
"We labelled, by hand, every voxel - the 3D equivalent to a pixel - with whether or not it was part of an aneurysm," said Chute, who is also co-lead author of the paper. "Building the training data was a pretty gruelling task and there were a lot of data."
The algorithm's conclusions are overlaid on the scan as a semi-transparent highlight, meaning medical personnel can still examine the underlying image. Eight clinicians tested HeadXNet by evaluating a set of 115 brain scans, once with the help of the AI and once without. With the tool, the clinicians correctly identified more aneurysms and therefore reduced the "miss" rate. The time it took to arrive at diagnosis was not affected.
Despite the early success, the Stanford team is cautious in relation to the clinical use of the tool. Different types of scanners or variations in patient population could impact the performance of the AI, so more work is needed before it is ready to detect brain aneurysms in the real world.
"Because of these issues, I think deployment will come faster not with pure AI automation, but instead with AI and radiologists collaborating," said co-senior author Andrew Ng, head of Stanford’s Machine Learning Group. "We still have technical and non-technical work to do, but we as a community will get there and AI-radiologist collaboration is the most promising path."