Multiple Organ Definition in CT Using a Bayesian Approach for 3D Model Fitting

Jennifer L. Boes, Ph.D. (1)
Terry E. Weymouth, Ph.D. (2)
Charles R. Meyer, Ph.D. (1)
(1) Department of Radiology, University of Michigan Medical Center
(2) Department of Electrical Engineering and Computer Science
The University of Michigan

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 set of biometric organ models--specifically for the liver and kidney--that accurately represents patient population shape information. Each model is generated by averaging surfaces from a learning set of organ 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 formulation of the model's priors and CT edge information, yielding a complete organ surface. We demonstrate this technique using a set of fifteen abdominal CT data sets for liver surface definition both before and after the addition of a kidney model to the fitting; we demonstrate the effectiveness of this tool for organ surface definition in this low-contrast domain.