This poster is published under an
open license. Please read the
disclaimer for further details.
Keywords:
Liver, Ultrasound, Intraoperative, Ablation procedures, Image registration
Authors:
J. Banerjee, C. Klink, E. D. Peters, W. J. Niessen, A. Moelker, T. V. Walsum; Rotterdam/NL
DOI:
10.1594/ecr2015/B-0856
Results
Thirteen 4D US volume sequences were acquired from six healthy volunteers at 6 Hz from an iU22 Philips machine.
From the 4D US sequences,
seven pairs of frames were selected in a systematic way such that they are representative of the whole breathing cycle.
Reference standard (ground truth) registration was established using the standard Elastix registration toolkit [1].
The Elastix toolbox is intensity based medical image registration application.
The Elastix results were visually inspected,
and as a result 6 pairs of images of the 91 pairs were excluded from the experiments.
Three pairs of images were excluded as they had large shadowing artefact,
poor quality of data and a large displacement between the frames.
Additional three pairs of images containing large vessel deformations were excluded for lack of uniqueness in rigid transformation.
For the rest of the 85 cases the Elastix registration performance was satisfactory.
For 85 pairs of 3D US volumes acquired from 4D US sequences,
a mean registration error of 1.3 mm is achieved.
A graphics processing unit (GPU) implementation runs the 3D US registration at 8 Hz.