An increased thoracic fat volume is associated with an increased risk of coronary artery disease, but epicardial fat is not routinely measured in clinical practice due to the time needed for manual quantification. As a result, the automation of epicardial adipose tissue quantification could affect the clinical routine and improve the risk assessment for asymptomatic patients, said Frédéric Commandeur, PhD, of Cedars-Sinai in Los Angeles.
Commandeur and colleagues at Cedars-Sinai have developed a deep-learning algorithm that can automatically segment and quantify thoracic fat volume on noncontrast CT images. Initial testing showed that the algorithm had excellent correlation with manual quantification performed by an expert. The technique is now being evaluated on multicenter and multireader data, he said.
"The demonstration of its efficiency on such a cohort would [enable the method] to move forward to a semiautomated or fully automated computer-aided diagnosis system to assist clinicians," he said. "Automated quantification could be performed directly after data acquisition and provide a robust result within seconds to improve risk assessment."
The availability of increasing computational power and access to larger amounts of data make it easier to train more-efficient models for making predictions, including for cardiac events, Commandeur added.
"We truly think computer-aided systems will have an important role to play in the next few years and clinicians will work in consensus with computers," he told AuntMinnie.com. "With the development of artificial intelligence, we [have] entered a very exciting and promising revolution [that] will lead to an improvement of patient care."
If you're still at RSNA 2017 on Friday, you can get a glimpse of this revolution by sitting in on this talk.
