AI tool screens CT scans for acute neurological illnesses
Researchers at Mount Sinai’s Icahn School of Medicine have developed an artificial intelligence tool that rapidly screens head CT scans.
Those who developed the system say it can be used to quickly review images for acute neurological illnesses, such as stroke and hemorrhage.
A deep neural network trained using 37,236 head CTs was tested in a randomized, double-blinded, prospective trial in a simulated clinical environment. Results of the study, published this week in the journal Nature Medicine, show that the AI algorithm was much faster than human diagnosis—in fact, 150 faster than the time it takes for physicians to read the images.
“With a total processing and interpretation time of 1.2 seconds, such a triage system can alert physicians to a critical finding that may otherwise remain in a queue for minutes to hours,” says Eric Oermann, MD, senior author and instructor in the Department of Neurosurgery at the Icahn School of Medicine.
According to the authors, their study is the first to leverage AI for detecting a wide range of acute neurologic events and to demonstrate a direct clinical application.
“We’ve married what computers do very well with a clinical need,” adds Oermann, who is a machine learning researcher and heads the Mount Sinai AI Consortium. “Computers will actually make medicine more efficient.”
“The expression ‘time is brain’ signifies that rapid response is critical in the treatment of acute neurological illnesses, so any tools that decrease time to diagnosis may lead to improved patient outcomes,” says Joshua Bederson, MD, co-author and professor and system chair for the Department of Neurosurgery at Mount Sinai Health System and clinical director of the Neurosurgery Simulation Core.
While the study conducted “weakly supervised classification” to screen CT images for acute neurologic events annotated with a “semi-supervised” natural language processing framework, going forward Oermann says researchers will focus on enhanced computer labeling of CT scans and a shift to “strongly supervised learning approaches” and novel techniques to increase data efficiency.
“Finding ways of improving the algorithm and tweaking it to make better use of medical data is one of our big scientific interests,” concludes Oermann.