Pearling: Medical Image Segmentation with Pearl Strings

This paper introduces a novel segmentation technique, called pearling, for identifying tubular structures in images. Examples of such structures include, but are not limited to, blood vessels, bones, roads, rivers, electrical wirings, and brush-strokes. Pearling allows the user to extract a higher level parametric representation of each tube or of a network of tubes. This representation is comprised of an ordered series of pearls, each defined by the location of its center, by its radius, and possibly by additional properties including a time value. A final model of the tube is then obtained by estimating continuous functions that interpolate the discrete series of pearls. Pearling is computationally efficient and well suited to user interactivity. We demonstrate its effectiveness in segmentation of medical images of different modalities as well as satellite imagery.