George Citroner, MedScape, writes that an artificial intelligence (AI) platform accurately identifies acute neurologic events, such as stroke, from CT scans in as little as 1.2 seconds, new research suggests.
If the findings are confirmed, this technology would radically speed the triage process by immediately alerting physicians to critical findings that may otherwise have remained in a queue from minutes to hours.
“This was our original study that launched our medical AI research consortium in the Mount Sinai Health System,” senior investigator Eric Oermann, MD, Department of Neurosurgery at the Icahn School of Medicine at Mount Sinai, New York City, told Medscape Medical News.
“I was motivated by my specific experiences with taking care of patients with acute neurologic illnesses where any possible way of reducing the time it took for me to reach them could have potentially improved their outcomes,” said Oermann.
The study was published online August 13 in Nature Medicine.
“Deep Learning” see link to study
Researchers used a dataset of 37,236 studies and an additional 96,303 radiology reports that were accumulated as part of the ICAHNC project, a computer vision initiative within the Department of Radiology and part of the Icahn School of Medicine AI Consortium (AISINAI).
The patients included in the test cohort had an average age of 59.7 years, and 51% were female. The studies were evenly distributed between acquisition settings, with 36% coming from the emergency department, 33% coming from an inpatient unit, and 31% coming from an outpatient setting.
The three most common symptoms at intake were headache, altered mental status, and ataxia/dizziness.
Images and the accompanying reports were standardized and processed by using a crowdsourcing platform and a natural language processing pipeline, a computer program that processes and analyzes large amounts of data using natural language.
The AI was then trained to interpret this information by using “deep learning,” a machine learning method that uses data representations instead of task-specific algorithms.
“This study was the perfect combination of my personal experiences as a neurosurgeon and my technical training as a mathematician and deep-learning researcher, and it’s the first of many projects that we expect to emerge from AISINAI in the near future,” Oermann said.
Both the natural language processing model and radiologists were assessed to find out how quickly they could recognize and provide notification of a critical finding.
A randomized controlled trial of interpretation speed in a simulated clinical environment showed that the AI platform could interpret CT scans 150 times faster than could a human radiologist.
Mount Sinai Health System