Description of shapes in CT images. The usefulness of time-series modeling techniques for identifying organs.

Abstract

An attempt has been made to investigate the usefulness of AR and CAR modeling techniques for shape description of substructures in CT images. The efficacy of the models have been checked using two organ shapes (liver and kidney) obtained from abdominal CT sections at nearly identical locations. The contours of the substructures representing the shapes, along with the sequence of coordinates, have been obtained by using edge detection followed by an edge-linking technique, where maximum and minimum lengths of the edge segments could be varied. The modeling techniques thus applied have resulted in feature vectors of the organ shapes corresponding to these models. Validity of the models has been assessed on the basis of lower values and consistency of Euclidean distances within a class, as well as the differences of these values with other classes.