Deep Learning Model for Coronary Artery Calcium Detection from Single-Shot Dual-Energy Chest X-rays: A Simulation Study

Cardiovascular disease (CVD) remains the leading cause of death in the United States, and early detection of coronary artery calcification (CAC) is crucial for preventive healthcare. While computed tomography (CT) is the gold standard for CAC assessment, its high cost and radiation burden limit widespread screening. Chest X-rays (CXR), being more accessible and lower in radiation, present an alternative for CAC detection. This study investigates the use of single-shot dual-energy (DE) CXR for CAC segmentation using a deep learning model based on nn-Unet architecture. We evaluated the model's performance across different CXR views and dose levels. Lateral bone images yielded the best CAC visibility with 81% sensitivity and 100% specificity for CAC detection, outperforming the anteroposterior (AP) view, which had only 46% sensitivity. Sensitivity decreased as dose levels were reduced, but specificity remained high. These results demonstrate the potential of DE-CXR as an effective tool for low-dose, accessible CAC detection.