Multispectral analysis of object surfaces extracted from volumetric data sets

This paper explores the potential of multispectral analysis of density slices extracted by automatic object recognition algorithms in volumetric data sets. The object recognition algorithm is utilized to define the object surface. Each point x on the object surface is hence associated with the set of measurements in the underlying volumetric data set that is within the region of interest and projected onto x. This variable size attribute vector is converted to a fixed size attribute vector by transforming the measurements into the associated cumulative distribution vector. The fixed size attribute vectors represent a multispectral image over the manifold defined by the object surface. This image is further processed with multispectral techniques such as the Karhunen-Loeve transform and visualized as a coloring of the 3-dimensional object surface. Our numerical experiments include 3-dimensional ultrasonic catheter studies of plaque formation in arteries, MRI studies of the brain, and evaluation of customized femoral hip prostheses based on CT imaging