Objectives: In US-guided radiofrequency ablation (RFA), multiple applications of RF are required for big liver tumors. Gas bubbles, or bleeding resulting from the first ablation may reduce visibility of tumors subsequently. Improving tumor localization during ablation benefit its efficacy. Image fusion of intra-intervention US with pre-intervention CT is a potential solution even though it is challenging due to respiration, pose position, and similarity measurement of US and CT. We proposed to use a learning-based registration of 3DCT with 2DUS to predict positions of tumors from visible vasculatures around them in the liver RFA. Learning-based
methods have not been reported on registering ultrasound and CT/MR images of the liver in the state-of-the-arts [1-4] so far.
Research supported by Agency for Science, Technology and Research, Singapore (grant number BEP 142 148 0022) and National Medical Research Council Singapore (NMRC/BnB/0017/2015) and (NMRC/