Characterization of Curve-Like Structures in 3D Medical Images
Gisela Klette, Mian Pan · 2005
This paper proposes algorithms for the analysis of sets of confocal microscope images of human brain tissue which constitute a 3D volume. The identification of suitable features for curve-like structures in these volumes is required for subsequent classification. The final goal is the distinction between brain tissues of patients with different degrees of neurological deceases. The given volumes show varying distributions, shapes and numbers of astrocytes (i.e., brain cells whose shape resembles that of a star). The hypothesis is that the “distribution ” of astrocytes in brain tissue is related to the number and distribution of branch nodes (clustered into junctions) in 3D skeletons of these cells. Segmentation is followed by an application of a modified 3D thinning algorithm. Further analysis is based on our definition of ‘junctions ’ and new methods for locating such junctions. We characterize their distribution based on subdividing the volume into subcubes, also using measures of complexity of junctions and distances between junctions (based on different metrics).