A team of researchers have developed an artificial intelligence-enabled tool they believe could make it easier to predict which patients are more likely to have a heart attack in the next 5 years. The Cedars Sinai-led group of investigators described the tool in The Lancet Digital Health, where they detailed their goal of developing and validating a deep learning system for coronary CT angiography (CCTA)-derived measures of plaque volume and stenosis severity (2022; https://doi.org/10.1016/S2589-7500(22)00022-X).
Describing CCTA as a "robust first-line test for the evaluation of coronary artery stenosis severity," the researchers pointed out that atherosclerotic plaque quantification from CCTA enables accurate assessment of coronary artery disease burden and prognosis. Beyond the assessment of stenosis severity, CCTA also enables non-invasive whole-heart quantification of atherosclerosis, according to the authors, adding that advances in CT technology also allow for semi-automated measurements of coronary atherosclerotic plaque with high accuracy when compared with intravascular ultrasound.
The algorithm the team has developed outlines coronary arteries in 3D images, and then identifies blood and plaque deposits in the coronary arteries, according to the investigators, who have found that the AI tool's measurements parallel the plaque amounts seen in coronary CT angiography.
Study Details
The researchers conducted an international, multi-center study, including nine cohorts of patients undergoing CCTA at 11 sites who were assigned to training and test sets. The team analyzed images from close to 1,200 patients who had undergone a coronary CTA at sites in Australia, Germany, Japan, Scotland, and the U.S., training their AI algorithm to segment and measure coronary plaque in 921 patients who had already been analyzed by doctors.
The deep learning network was then applied to an independent test set, which included an external validation cohort of 175 patients (1,081 lesions) and 50 patients (84 lesions) assessed by intravascular ultrasound within 1 month of CCTA, wrote the researchers, who evaluated the prognostic value of deep learning-based plaque measurements for fatal or non-fatal myocardial infarction in 1,611 patients from the prospective SCOT-HEART trial, assessed as dichotomous variables using multivariable Cox regression analysis, with adjustment for the ASSIGN clinical risk score.
Within the overall test set, the researchers saw "excellent or good agreement," respectively, between deep learning and expert reader measurements of total plaque volume. When compared with intravascular ultrasound, there was "excellent agreement for deep learning total plaque volume and minimal luminal area, the authors wrote, adding that myocardial infarction occurred in 41 of 1,611 patients from the SCOT-HEART trial over a median follow-up of 4-7 years.
"Heart attacks are very unpredictable, but can have devastating if not fatal consequences," stated David Newby, MD, Chair of Radiology at the University of Edinburgh, and a co-author of the study. "Using a modern CT scanner, we can now see inside the heart arteries and measure the amount and the type of disease. We have previously shown that this is the best way of predicting a future heart attack."
Accurate measurement of this disease requires a good deal of training, however, and is very time-consuming and laborious, taking at least 30 minutes per scan, he noted. "We hypothesized that deep-learning plaque analysis could automate this process in a rapid (6 seconds per scan), reproducible, and accurate manner to predict future heart attacks."
The prediction of future heart attacks is based on the principle that the disease in the heart arteries "is the substrate for a heart attack," Newby continued. "This is true not only for the total amount of plaque, but also for the type of plaque. The risk of a heart attack is particularly high when the arteries have a lot of soft non-calcified plaques that are rich in lipids and cholesterol. These types of plaque can be seen on CT and detected by the algorithm."
Newby reasons that the automated approach he and his colleagues took in this study could ultimately help greatly in the prediction of future heart attacks for all clinicians, "and perhaps prompt them to commence treatment with cardiovascular preventive therapies, such as statins and aspirin.
"This will be of assistance to oncologists who use cancer treatments that are known to be associated with an increased risk of a heart attack, such as 5-fluorouracil-based regimens," Newby concluded. "This will hopefully enable them to improve selection of therapies for their patients and hopefully avoid the precipitation of a heart attack."
Mark McGraw is a contributing writer.