Thoracic Aortic Calcification (TAC) is a significant predictor of cardiovascular events, but it's often overlooked in routine CT scans. This study explores the use of these non-dedicated scans for opportunistic TAC detection and quantification. Our model was trained and validated on a dataset comprising 661 chest CT exams retrospectively collected from patients undergoing imaging for clinical indications. TAC regions were annotated by experts to serve as ground truth. We developed a deep learning model based on an enhanced 3D U-Net with residual blocks to automatically segment TAC and predict the Agatston calcium score. Using expert annotations as a reference, our model achieved strong performance, with a Pearson's correlation coefficient (ρ) of 0.87 ± 0.06 and a coefficient of determination (R²) of 0.76 ± 0.11. Our work introduces a compact and efficient deep learning model for the automated segmentation and quantification of the thoracic aortic calcium score in routinely acquired images. These findings support the feasibility of applying deep learning for reliable, automated assessment of TAC in routine CT exams, without requiring dedicated cardiac imaging. By transforming incidental data into clinically relevant information, this approach enables proactive cardiovascular risk assessment, with the potential to improve patient outcomes and enhance healthcare efficiency.