Calcification of the coronary arteries is a significant and independent predictor of adverse cardiovascular events such as heart attacks. Nonetheless, despite this knowledge and the fact that it can be assessed using any chest CT scan, quantification of coronary artery calcium (CAC) is not automatically integrated into the patient pathway because it requires radiological expertise, time, and specialized equipment.
To address this issue, a multidisciplinary team developed and tested a deep-learning algorithm that can automatically quantify CAC from any chest CT scan. The deep-learning system, in theory, does much of what a human would do to quantify calcium.
To train the deep-learning system, the researchers used manual segmentations performed by expert CT readers on 1636 scans from a seminal yet ongoing cardiovascular study to identify and quantify CAC. In less than 2 seconds, the deep-learning system predicts the heart center, segments the heart, and segments and identifies coronary calcium using three consecutive convolutional neural networks. The CAC scores are then computed and classified into clinically relevant categories: very low (CAC=0), low (CAC=1-100), moderate (CAC=101-300), and high (CAC>300).
The study’s strength is the breadth and scope of the datasets on which the deep-learning system was later tested. The researchers used four cohorts with different pathologies: 663 Framingham Heart Study participants (none of whom were in the training group) had cardiac CT; 14,959 heavy smokers had lung cancer screening CT (NLST trial); 4021 patients with stable chest pain had cardiac CT (PROMISE trial); and 441 patients with acute chest pain had cardiac CT (ROMICAT-II trial).
Related Content: Optical Fiber Imaging For Medical Applications