Tracking liver motion using 3-D ultrasound and a surface based statistical shape model

Abstract

We present a technique for registering information from preoperative CT or MR images to physical space using intraoperatively acquired 3-D ultrasound data and a surface-based statistical shape model. The model is subject-specific and captures the statistical modes of variation of the liver surface through the breathing cycle. The registration uses a Bayesian formulation, which enables information about the likely position in the breathing cycle to be incorporated in the form of prior knowledge. It is computed using the model and the ultrasound image intensities, and is constrained by the model to produce 'realistic' surfaces. Once an initial registration is computed, the liver motion and deformation can be tracked using a single ultrasound image combined with the statistical model. The technique is demonstrated by registering models constructed for 3 different volunteers to ultrasound data acquired at different points in the breathing cycle. This method has potential application in treatment of any abdominal organ which is affected by breathing motion.