Cardiovascular disease (CVD) is the leading cause of death worldwide. About 610,000 people die of heart attacks and strokes in the U.S. every year, according to the Center for Disease Control and Prevention, and worldwide, the number stands at about 17.9 million. CVD isn’t impossible to predict, fortunately — there’s a strong risk factor in coronary artery calcium (CAC) deposits that restrict blood flow. Unfortunately, measuring CAC requires experts who can closely inspect computerized tomography (CT) scans for worsening signs and symptoms.

But there’s hope yet for a more automated approach.

A newly published paper on the preprint server Arxiv.org (“Direct Automatic Coronary Calcium Scoring in Cardiac and Chest CT“) proposes an artificially intelligent (AI) system that can evaluate and score CAC without human supervision. That’s not especially novel — automated CAC tests have been around for a while. However, the coauthors claim that their system is up to hundreds of times faster than state-of-the-art methods.

“Current automatic calcium scoring methods are relatively computationally expensive and only provide scores for one type of CT,” they explain. “[Our] method achieves robust and accurate predictions of calcium scores in real-time.”

The researchers’ AI system comprises two convolutional neural networks, a class of deep neural networks commonly applied to analyzing visual imagery. The first takes as input CT scans and aligns the fields of view, and the second performs direct regression — i.e., linear modeling of the relationship between variables — of the calcium score.

The networks were trained on two datasets: one from the University Medical Center Utrecht in the Netherlands containing 903 cardiac CT scans, of which 237 scans were used for training; and 1,687 chest CT scans from the National Lung Screening Trial (1,012 of which were used for training). In experiments conducted on an Intel-based PC with an Nvidia Titan X graphics card, the AI algorithms predicted calcium scores in less than 0.3 seconds, with a correlation coefficient (a measure of strength between two variables, in this case between predicted and manual calcium scores) of 0.98 for both cardiac and chest CT scans.

The new paper comes months after researchers at Florida State University and the University of Florida, Gainesville detailed an AI system that could predict one-year mortality in ICU patients who’d experienced a heart attack, and after Corti, an AI system which detects heart attacks during emergency phone calls, started rolling out to London, Paris, Milan, and Munich. It also follows on the heels of Zebra Medical’s successful bid to obtain FDA 510(k) clearance for its coronary calcium scoring algorithm.