Bayesian vessel extraction for planning of radiofrequency-ablation

Abstract

The software-assisted planning of radiofrequency-ablation of liver tumors calls for robust and fast methods to segment the tumor and surrounding vascular structures from clinical data to allow a numerical estimation, whether a complete thermal destruction of the tumor is feasible taking the cooling effect of the vessels into account. As the clinical workflow in radiofrequency-ablation does not allow for time consuming planning procedures, the implementation of robust and fast segmentation algorithms is critical in building a streamlined software application tailored to the clinical needs. To suppress typical artifacts in clinical CT or MRT data - like inhomogeneous background density due to the imaging procedure - a Bayesian background compensation is developed, which subsequently allows a robust segmentation of the vessels by fast threshold based algorithms. The presented Bayesian background compensation has proven to handle a wide range of image perturbances in MRT and CT data and leads to a fast and reliable identification of vascular structures in clinical data.