Machine learning can help physicians exclude the presence of obstructive coronary artery disease (CAD) in patients once they undergo coronary artery calcium scoring (CACS), according to new research published in Academic Radiology. Could this help limit unnecessary imaging examinations?
CACS is a “low-dose, reliable technique” for estimating a patient’s atherosclerotic coronary plaque burden, the authors explained, and coronary artery calcium (CAC) is associated with a variety of cardiovascular events. The researchers aimed to use AI to see if combining CAC, aortic root calcification (ARC) and thoracic aortic calcification (TAC) estimations could help reveal data related to CAD.
The team’s machine learning-based algorithm was built with data from 435 patients who underwent a clinically-indicated CACS test from September 2017 to July 2018. The cohort was 77% female with a mean age of 61 years old, and all patients had a “low to moderate probability of CAD.” The gradient boosting machine (GBM) model was then tested on data from 126 consecutive patients who underwent a CACS test from September to October 2018.
Overall, when evaluating the 126-patient cohort, the AI model’s results included 73 true negatives, 0 false negatives, 20 true positives and 33 false positives. This means that 73 coronary CT angiography (CCTA) examinations—which are associated with both radiation exposure and a risk of contrast-induced renal failure—could have potentially been avoided altogether.
“Application of the proposed method can significantly decrease the number of patients referred to CCTA (by about 70%), which leads to the limitation of patients’ exposure to side effects of such tests and can reduce the costs of diagnostics for medical care providers,” wrote lead author Jordina Torrents-Barrena, PhD, Universitat Pompeu Fabra in Barcelona, Spain, and colleagues.
The authors did note their study had certain limitations. It was a single-center study, for instance, and the study population “was relatively healthy,” with obstructive CAD being observed in just 16% of patients from the control group.
Michael has more than 16 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.