Using a Deep Learning System to Identify Brain Aneurysm

Using a Deep Learning System to Identify Brain Aneurysms

New artificial intelligence (AI) tools could enable doctors to detect brain aneurysms before it is too late.

Researchers from Stanford University have developed a new tool built around an algorithm dubbed HeadXNet that pinpoints areas within a brain scan that are more likely to contain an aneurysm—bulges in blood vessels in the brain that can leak or burst open, potentially leading to stroke, brain damage or death.

HeadXNet allowed clinicians to correctly identify aneurysms at a level equivalent to finding six extra diagnoses per 100 brain scans.

“There's been a lot of concern about how machine learning will actually work within the medical field,” Allison Park, a Stanford graduate student in statistics and co-lead author of the paper, said in a statement. “This research is an example of how humans stay involved in the diagnostic process, aided by an artificial intelligence tool.”

The researchers trained the algorithm with 611 computerized tomography (CT) angiogram head scans, allowing the tool to decide in each voxel of a scan whether an aneurysm is present.

“We labelled, by hand, every voxel—the 3D equivalent to a pixel—with whether or not it was part of an aneurysm,” Christopher Chute, a graduate student in computer science, who is also co-lead author of the paper, said in a statement. “Building the training data was a pretty grueling task and there were a lot of data.”

When an aneurysm is present, the algorithm will overlay a semi-transparent highlight on top of the scan.

“We were interested how these scans with AI-added overlays would improve the performance of clinicians,” Pranav Rajpurkar, a graduate student in computer science and co-lead author of the paper, said in a statement. “Rather than just having the algorithm say that a scan contained an aneurysm, we were able to bring the exact locations of the aneurysms to the clinician's attention.”

One of the benefits of using AI for brain scans is researchers will be able to scroll through hundreds of images quickly and gain a better grasp of potential problems. This is particularly important because aneurysm come in different sizes and shapes and can balloon out at unusual angles, registering a small blip within a movie-like succession of images.

“Search for an aneurysm is one of the most labor-intensive and critical tasks radiologists undertake,” Kristen Yeom, associate professor of radiology and co-senior author of the paper, said in a statement. “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.”

The new tool will also improve the consensus when multiple clinicians are interpreting a brain scan.

To test the new system, a team of eight clinicians evaluated a set of 115 brain scans both with and without the help of the AI tool and found they were able to find more aneurysms using the tool.

However, HeadXNet did not speed up the researcher’s ability to decide on a diagnosis or correctly identify scans without aneurysm.  

HeadXNet was also tested during the AI for Healthcare Bootcamp, which was organized by Stanford's Machine Learning Group. The researchers were tasked with creating an AI tool that accurately processes large stacks of 3D images and complements clinical diagnostic practices.

The researchers plan to investigate HeadXNet further in a multi-center collaboration to evaluate generalizability before they are able to deploy it in a clinical setting.  

They also believe the underlying machine learning techniques could be used to diagnose diseases both inside and outside of the brain.

According to the researchers, intracranial aneurysm occur in between 1 and 3 percent of the population, accounting for more than 80 percent of the non-traumatic life-threatening subarachnoid hemorrhages.

The study was published in JAMA Network Open.

By

Vengatajalapathi veerapan