Liver definition in CT using a population-based shape model

Abstract

Organ definition in computed tomography (CT) is of interest for treatment planning and response monitoring. We present a method for organ definition using a priori information about shape encoded in a biometric organ model--specifically a liver model--that accurately represents patient population shape information. This model is generated by averaging surfaces from a learning set of liver shapes previously registered into a standard space defined by a small set of landmarks. The model is placed in a specific patient's data set by identifying these landmarks and using them as the basis for model deformation; this preliminary representation is then iteratively fit to the patient's data based on a Bayesian combination of the model's priors and CT edge information, yielding a complete organ surface. We demonstrate this technique on a set of ten abdominal CT data sets and show its effectiveness as a tool for organ surface definition in this low-contrast domain.