Centerline Detection and Pericoronary Adipose Tissue Prediction Using Image-to-Image Regression in Coronary CTA Images

Automated coronary artery centerline detection in coronary computed tomography angiography (CCTA) images is critical for subsequent analyses of structures including pericoronary adipose tissue (PCAT), an indicator of cardiovascular risk. We developed methods for coronary centerline detection using 3D image-to-image regression, utilizing a dataset of 83 CCTA volumes. Image-to-image regression was used as it may be superior to vessel segmentation in ensuring vessel continuity in cases with severe stenoses. We compared PCAT assessments obtained from our automated centerline regression method to those from ground truth centerlines obtained with highly interactive clinical software. For PCAT volume and mean PCAT HU value, there was an insignificant difference (p≥0.03) between each metric obtained from the ground truth and predicted centerlines for the LCX, LAD, and RCA. We report a 1.56% difference in mean PCAT HU and a 12.8% difference in mean PCAT volume, both within clinically reasonable ranges, demonstrating the utility of this method for automated PCAT analysis.