[ad_1]
Researchers at Stanford University in the United States have developed an artificial intelligence (AI) tool to better detect brain aneurysms.
Based on an algorithm called HeadXNet, the new tool is designed to highlight areas of a brain scan that may contain aneurysms, likely to leak or open up and possibly causing a stroke, brain damage or death.
The researchers found that the AI solution improved the ability of clinicians to accurately detect aneurysms and a consensus among clinicians.
Kristen Yeom, badociate professor of radiology at Stanford University, said, "The search for an aneurysm is one of the most critical and arduous tasks that radiologists must undertake.
"Given the challenges inherent in complex neurovascular anatomy and the potential fatal outcome of a forgotten aneurysm, this has prompted me to apply advances in computer science and to vision to neuroimaging. "
The team trained the AI algorithm using clinically significant aneurysms detectable from 611 CT scan head angiography scanners.
The solution badyzes each voxel for a scan to find the presence of an aneurysm and generates its findings in the form of semi-transparent highlighting.
Pranav Rajpurkar, a graduate student in computer science from Stanford University, said, "Rather than asking the algorithm to say that a scan contained an aneurysm, we were able to accurately pinpoint the locations of the aneurysms. the attention of the clinician.
"Searching for an aneurysm is one of the most critical and critical tasks that radiologists must undertake."
When eight clinicians evaluated HeadXNet by examining 115 brain scans in search of aneurysm, the tool allowed the correct identification of more aneurysms, thus minimizing the rate of misfires.
However, further studies are needed to determine the generalizability of the algorithm before clinical application in different hospital centers. To solve this problem, the team plans a multicenter alliance.
HeadXNet's machine learning technique should also be trained to detect other diseases.
Source link
