A data generation pipeline for cardiac vessel segmentation and motion artifact grading

Cardiac CT is a first-line imaging modality for the diagnosis and monitoring of cardiovascular disease. A major challenge of cardiac CT remains its vulnerability to motion artifacts due to fast and/or irregular cardiac dynamics. Existing motion artifact suppression algorithms may be improved by accounting for distribution shifts due to natural and pathological variations in anatomic presentation, as well as the effects on image appearance due to specific scan protocols and scanner hardware selection. In this paper, we construct a dataset containing over 1,000 cardiac CT images that is enriched with diverse features in order to improve model generalization capabilities. In addition, we provide a pipeline for source-agnostic vessel segmentation and motion artifact scoring. We demonstrate the merits of the approach and provide a guideline for ensuring source-agnostic representation of anatomical and pathological imaging biomarkers in cardiac CT applications that may serve as a template for other clinical applications.