Publication Details
Fotios Drakopoulos and Nikos Chrisochoides.
Published in VMASC 2014 Capstone Conference, Best paper in Medical Simulations, April, 2014
Abstract
We present a Parallel Adaptive Physics-Based Non- Rigid Registration (PAPBNRR) framework for warping pre-operative to intra-operative brain Magnetic Resonance Images (MRI) of patients who have undergone a tumor resection. This method extends our previous APBNRR framework based on ITK and contributes three new parallel modules: A Finite Element Method (FEM) Solver for assembling the system matrices of a heterogeneous biomechanical model, rejecting the feature outliers and estimating the model deformations; two correction modules, the first for the warped pre-operative segmented MRI and the second for the produced image deformation field, which take into account the tissue removal depicted in the iMRI. As a result, PAPBNRR not only accurately captures the large intra-operative deformations associated with the resection, but also further reduces the overheads due to the inherited-from APBNRR- adaptivity and brings the end-to-end execution within the time constraints imposed by the neurosurgical procedure. Our evaluation is based on 6 clinical volume MRI cases including: (i) partial and complete tumor resections, (ii) isotropic and anisotropic image spacings. In all the case studies, the PAPBNRR framework shows promising results. In a Linux Dell workstation with 12 Intel Xeon 3.47 GHz CPU cores and 96GB of RAM, it reduces the registration time from about 10 minutes to less than 140 seconds for the anisotropic data and from 2 minutes to less than 45 seconds for the isotropic data.
