Reducing Artifacts Impact on IVUS Automatic Segmentation Via Inpainting

Abstract

In this work we present a novel approach that uses digital inpainting to preprocess intravascular ultrasound (IVUS) images to reduce the impact of undesired features. Then, we automatically segment the arterial wall with active contour models. IVUS is a catheter-based medical imaging technique that produces cross-sectional images of blood vessels. Segmentation of vessel wall is particularly useful to study many coronary artery diseases, such atherosclerosis. Being IVUS a good technology to analyse the anatomy of the arterial wall, the modality may present several artifacts, such as shadows or catheter ring-down, that may difficult further processing. To deal with these artifacts, in this paper we consider an exemplar-oriented inpainting algorithm that replaces the corrupted information by using the unaltered neighbourhood. To determine the impact of this preprocessing step, segmentation results over inpainted and non-inpainted IVUS are presented. The images are compared with manually outlined contours, showing that the inpainting method promotes continuity of the arterial wall and improves the segmentation performance.