Analytics in Action News

Michigan Medicine Uses AI to Measure Aortic Growth

An artificial intelligence-based method for measuring aortic growth promotes early invention for fatal heart conditions, according to researchers.

artificial intelligence

Source: Getty Images

By Erin McNemar, MPA

- Michigan Medicine researchers have developed a new artificial intelligence method of measuring aortic growth that can assist clinicians in identifying potentially fatal heart conditions earlier.

The vascular deformation mapping technique measures changes in the thoracic aorta, which carries blood from the heart to the rest of the body.  The new method uses high-resolution CT imaging and AI to calculate three-dimensional changes in the aortic wall. According to researchers, the AI method significantly outperforms the standard manual rating methods done by experts.

“The technique used in this algorithm has been around for a while, but no one has ever used it to see three-dimensional growth of an aneurysm of the thoracic aorta,” Nicholas Burris, MD, corresponding author of the paper and assistant professor of radiology at Michigan Medicine, said in a press release. “This is a promising step towards having technology that pushes the accuracy of measurement past what human raters can achieve, allowing clinicians to have the best possible picture of a patient’s condition.”

A thoracic aortic aneurysm occurs when the largest part of the aorta becomes weakened and grows, or dilates, increasing the risk of a potentially fatal rupture or dissection. Physicians recommend regular testing and CT scans to measure aortic growth for about 3 percent of adult patients over 50 with this largely asymptomatic condition.

The current standard practice to measure growth is done with human “rates” who line up two images and draw a line at two points to find the change. However, this process is prone to error, according to Burris, the director of aortic imaging at Michigan Medicine. Doctors cannot confidently tell if the thoracic aorta grows in many cases, creating uncertainty regarding the best treatment plan.

READ MORE: Researchers Study Artificial Intelligence Tool to Detect Esophageal Cancer

“The challenge we’re faced with clinically is that a typical aneurysm in the aorta is going to grow only a fraction of a millimeter every year, and the process of manually drawing diameters that precise is very hard to reproduce,” Burris said. “You have a lot of variability in standard measurements relative to a minimal amount of actual aneurysm growth. Basically, you rarely end up getting a confident assessment of growth, which can make it difficult to know what the patient’s actual risk is and how closely they need to be followed with repeat CT scans.”

The vascular deformation mapping method created by Burris’ team relies on an image analysis technique known as image registration. Image registration aligns the anatomy shown in multiple CT scans by taking any pixel on the first scan and comparing its exact position to the pixel on the second scan.

When all the pixels are aligned, a three-dimensional color map of the aorta shows how much and where the thoracic aorta has grown.

The research team used scans from almost 50 aortic aneurysm patients and 75 reference models with variable growth of the aortic wall. They tested the automated program against two expert manual raters and found the vascular deformation mapping technique outperformed the humans with higher accuracy and lower variability in the growth measurements.

“Recent advances in artificial intelligence have generated a lot of interest in AI in relation to automating radiology tasks,” said Charles Hatt, PhD, co-author of the paper and adjunct research assistant professor of radiology at Michigan Medicine, in the news release. “It turns out that replacing human radiologists is not a simple task, and a more realistic goal for AI is to speed-up workflows and assist radiologists in making object, quantitative measurements that are otherwise cumbersome to perform and inherently subjective. In this regard, vascular deformation mapping is a perfect, real-world example of how AI and quantitative imaging can improve clinical care by empowering clinicians rather than attempting to replace them.”

READ MORE: Artificial Intelligence Can Help Free Up Anesthesiologists’ Time in OR

While reports indicate that vascular deformation mapping may be more beneficial than the human rating to inform whether surgery is necessary on an aortic aneurysm, researchers explained that the approach must be studied further in large groups of patients in a clinic.

Fortunately, the vascular deformation mapping technique can be performed on routine CT scans of the aorta, making it easy to conduct extensive research studies.

“This is a totally new way of looking at aortic aneurysm growth,” Burris said. “As this develops, there is a possibility to deploy this across a larger spectrum of diseases, such as abdominal aortic aneurysm.  Moving from the current one-dimensional measurements to a three-dimensional approach lets us see patterns of aneurysm growth in a way never before possible and allows us to ask many new questions and learn how a highly-accurate tracking tool like this can be used to ultimately improve care for patients.”