Detection and Quantification of Line and Sheet Structures in 3-D Images

Abstract

This paper describes a method for the accurate segmentation of line and sheet structures in 3-D images with the aim of quantitatively evaluating anatomical structures in medical images with line-like shapes such as blood vessels and sheet-like shapes such as articular cartilages. Explicit 3-D line and sheet models are utilized. The line model is characterized by medial axes associated with variable cross-sections, and the sheet model by medial surfaces with spatially variable widths. The method unifies segmentation, model recovery, and quantification to obtain 3-D line and sheet models by fully utilizing formal analyses of 3-D local intensity structures. The local shapes of these structures are recovered and quantitated with subvoxel resolution using spatially variable directional derivatives based on moving frames determined by the extracted medial axes and surfaces. The medial axis detection performance and accuracy limits of the quantification are evaluated using synthesized images. The clinical utility of the method is demonstrated through experiments in bronchial airway diameter estimation from 3-D computed tomography (CT) and cartilage thickness determination from magnetic resonance (MR) images.