Automatic extraction of arterial centerline from whole-body computed tomography angiographic datasets

The aim of this study was to develop a new, fully-automated approach for the extraction of arterial centerlines from whole-body computed tomography angiographic (CTA) data sets for the planning of transcatheter aortic valve replacement (TAVR) procedures by the transfemoral approach. The method starts with an image pre-processing step to correct contrast inhomogeneities in different image slices, followed by the detection of anatomical landmarks and subsequent wave propagation, generating a cost matrix that represents the wave traveling time. Finally, the arterial centerlines are extracted from the aortic root to the femoral arteries using the conventional Dijkstra algorithm. 36 patients who underwent TAVR procedures with whole-body CTA scans were studied; 15 for training and the remaining 21 for testing. In 1 training case and 2 testing cases, user interaction was needed to correct the centerline (in the femoral artery). In the remaining 33 cases (91.7%), a fully-automated centerline was obtained with excellent results. In all cases, the average root mean square error was 2.55±0.70mm (training 2.26±0.48mm versus testing 2.77±0.77mm, p = 0.03) and the average mean error was 1.63±0.40mm (training 1.46±0.35mm versus testing 1.74±0.39mm, p = 0.02).