Deformable Registration of MRA and 4D Flow Images to Facilitate the Estimation of Flow Properties within Large Blood Vessels

Abstract

A method is presented for registering MRA and 4D Flow images of large blood vessels (e.g., aorta, brachiocephalic, left common carotid, and left subclavian) with the aim of combining information from both modalities to calculate blood flow properties within these vessels. In particular, the vessel surface segmented from an MRA image is aligned with the blood velocity field extracted from a corresponding 4D Flow image. The registration algorithm is driven by aligning vessel centerlines extracted from the two images. The proposed method is robust to small deformations, partial omissions of vessel surfaces, and noise in the blood velocity field. It is tested on image pairs from 8 patients with single-ventricle physiology. The quality of the resulting alignment is assessed using (i) histograms of distances between corresponding vessel centerlines, (ii) estimates of blood flow using the aligned and unaligned segmentations (as extracted from the MRA image), and (iii) visual comparisons of the aligned and unaligned segmentations together with the (possibly deficient) segmentation extracted from the corresponding 4D Flow image. Following registration, medians of the distance histograms between corresponding centerlines decrease by an average of 80.6%, blood flow estimates at systole increase at all three measured locations along the aorta for every patient, and visualizations show a significant improvement in the alignment between corresponding segmentations.