Deep learning tool could predict future myocardial infarction

Damini Dey

Researchers have detailed the use of an artificial intelligence-enabled tool for atherosclerotic plaque quantification from coronary CT angiography (CCTA), which they claim may make it easier to predict if a person will have a heart attack.

The tool, described in a paper published in The Lancet Digital Health by Andrew Lin, Damini Dey (both Cedars-Sinai Medical Center, Los Angeles) et al provides rapid measurements of plaque volume and stenosis severity from CCTA that agree closely with expert readers and intravascular ultrasound (IVUS), and could have prognostic value for future myocardial infarction (MI) the researchers suggest.

“Coronary plaque is often not measured because there is not a fully automated way to do it,” said Dey, senior author of the study. “When it is measured, it takes an expert at least 25 to 30 minutes, but now we can use this program to quantify plaque from CTA images in five to six seconds.”

Dey and colleagues analysed CTA images from 1,196 people who underwent a coronary CTA at 11 sites in Australia, Germany, Japan, Scotland and the USA. The investigators trained the AI algorithm to measure plaque by having it learn from coronary CTA images, from 921 people, that already had been analysed by trained doctors.

The algorithm works by first outlining the coronary arteries in 3D images, then identifying the blood and plaque deposits within the coronary arteries. Investigators found the tool’s measurements corresponded with plaque amounts seen in coronary CTAs. They also matched results with images taken by two invasive tests considered to be highly accurate in assessing coronary artery plaque and narrowing: IVUS and catheter-based coronary angiography.

Finally, the investigators discovered that measurements made by the AI algorithm from CTA images accurately predicted heart attack risk within five years for 1,611 people who were part of a multicentre trial, the SCOT-HEART trial.

In their Lancet Digital Health paper, Lin, Dey, et al note that in the overall test set, there was excellent or good agreement, respectively, between deep learning and expert reader measurements of total plaque volume (intraclass correlation coefficient [ICC] 0.964) and percent diameter stenosis (ICC 0.879; both p<0.0001).

Furthermore, when compared with IVUS, there was excellent agreement for deep learning total plaque volume (ICC 0·949) and minimal luminal area (ICC 0.904). The mean per-patient deep learning plaque analysis time was 5.65 s (SD 1.87) versus 25.66 min (6.79) taken by experts.

Over a median follow-up of 4.7 years (IQR 4–5.7), MI occurred in 41 (2.5%) of 1,611 patients from the SCOT-HEART trial. A deep learning-based total plaque volume of 238.5 mm3 or higher was associated with an increased risk of MI (hazard ratio [HR] 5.36, 95% CI 1.70–16.86; p=0.0042) after adjustment for the presence of deep learning-based obstructive stenosis (HR 2.49, 1.07–5.50; p=0.0089) and the ASSIGN clinical risk score (HR 1.01, 0.99–1.04; p=0.35).

“More studies are needed, but it is possible we may be able to predict if and how soon a person is likely to have a heart attack based on the amount and composition of the plaque imaged with this standard test,” said Dey.

Detailing some limitations of the study, the authors note that data on patient race or ethnicity were not uniformly available for all sites; however, they add that the training dataset drew from diverse and geographically distinct populations.  A further limitation offered by the study team is that although they showed the robust performance of deep learning across several different CT vendors and scan parameters, they excluded CCTA studies of poor image quality that were deemed uninterpretable by expert readers.

“Despite these limitations, our study represents the first validation of a deep learning approach for atherosclerotic quantification from CCTA using invasive reference standards, and is the first demonstration of the predictive value of deep learning-based plaque measurements for risk of cardiac events,” they write.

Dey and colleagues are continuing to study how well their AI algorithm quantifies plaque deposits in patients who undergo coronary CTA.


Please enter your comment!
Please enter your name here