The Duke Division of Cardiology is using big data to help physicians analyze images from coronary angiograms more quickly and accurately.
A team led by Manesh Patel, MD, chief of the Division of Cardiology and co-director of Duke Heart, is creating computer algorithms that can analyze angiogram images in real time to identify significant blockages and abnormalities and help guide treatment decisions.
“The computer can look at every single pixel of a digital angiogram (or an echocardiogram, or CT scan)—megabytes or gigabytes of data,” Patel notes. “It can pick up things that our eyes may not be able to see or our brains process, even with experience. When you need to make quick decisions, especially when a patient is unstable, that information is invaluable.”
In Project Deep Angio, Patel and co-investigator Ricardo Henao, PhD, an assistant professor in biostatistics and bioinformatics at Duke’s Pratt School of Engineering, teamed up with imaging company Image Share (founded by a Pratt graduate) to calibrate and measure stenosis, or blood vessel narrowing, on over 1,000 angiograms from Duke patients. Then they trained the computer to read each angiogram to assess the severity of each blockage.
“With the first thousand angiograms, we were able to get the computer’s readings to about 82 percent as accurate as the physician’s eyeballing” says Patel. “Now we’re training the computer with another couple of thousand angiograms, so it will keep learning and becoming more accurate.”
Eventually, Patel foresees, these algorithms will be incorporated into the angiogram software and machines around the world to help physicians analyze images consistently and in real time. Data patient outcomes can also be added, so computers can learn to identify which images are associated with future heart attacks, for example.
“With the revolution in computing power and digitalized information, we have an amazing opportunity to bend the curve of cardiovascular and human health to improve outcomes,” Patel says. “In areas where we, as practitioners, touch and interact with patients and make decisions, this technology can be supportive and efficiency-driving—to complement personal care, not replace it.”